--- path: manual/index.md title: Stevin Manual summary: How to use Stevin: get a workspace, connect the places you work, hand it tasks, and control what it may do. --- ## What Stevin is Stevin is an AI colleague. It lives in the chat tools your team uses every day, in a web console, and on the phone. You hand it a piece of work; it looks things up and takes action in your connected tools, stops to ask before any action that needs a person's sign-off, and sends the result back when it is done. For a fuller introduction, see [What is Stevin](get-started/what-is-stevin.md). ## Where to start | Who you are | Read first | |---|---| | New to Stevin | [Get started](get-started/index.md) | | Rolling Stevin out to your team | [Get a workspace](get-started/get-a-workspace.md), then [Channels](channels/index.md) and [Integrations](customize/integrations.md) | | An everyday member | [Your first conversation](get-started/first-conversation.md), then [Common uses](get-started/use-cases.md) | | An admin | [Customize](customize/index.md): integrations, MCP servers, skills, people, permissions, usage | | Concerned with security and compliance | [Security and data](reference/security-and-data.md) | | Stuck on a problem | [Troubleshooting](troubleshooting/index.md) | | Building an AI agent that needs this manual | [Use this manual from an AI agent](reference/manual-for-agents.md) | ## How the manual is organized - **[Get started](get-started/index.md)**: what Stevin is, getting a workspace, signing in, your first conversation. - **[Channels](channels/index.md)**: where you talk to Stevin. Slack, Feishu and Lark, WhatsApp, the console, live voice and the phone. - **[Features](features/index.md)**: what Stevin does. Tasks, approvals, automations, skills, memory, the library, code execution. - **[Customize](customize/index.md)**: everything you set up in the console. Integrations, MCP servers, skills, default models, people and roles, permissions, usage and plans, your own settings. - **[Troubleshooting](troubleshooting/index.md)**: what to check when something does not work. - **[Reference](reference/index.md)**: glossary, the full permission list, chat commands, security and data. ## Conventions - **Bold** names are things you click or see in the console, such as **Activity** or **Customize**. - Screenshots carry numbered callouts. The list under each screenshot says what every number points at, so the page reads the same with or without the image. - Steps that only an admin can do say so. Roles are explained in [People and roles](customize/people-and-roles.md). ## Versions and feedback The manual describes the version of Stevin that is currently live. Each page shows when it was last updated. If something here does not match the product, write to [support@stevinhq.com](mailto:support@stevinhq.com). --- path: manual/get-started/index.md title: Get started summary: Four steps to start using Stevin: learn what it is, get a workspace, connect a channel, and have your first conversation. --- ## Four steps 1. [What is Stevin](what-is-stevin.md): two minutes on what it can and cannot do. 2. [Get a workspace and sign in](get-a-workspace.md): request a workspace and sign in to the console. 3. Pick a [channel](../channels/index.md) to add Stevin to the chat tool your team already uses, then connect [integrations](../customize/integrations.md) as needed. 4. [Your first conversation](first-conversation.md): how to hand over a task and how to see what Stevin did. ## What you need - A work email address, used to request the workspace and to sign in. - Admin rights in your chat tool (or access to someone who has them), to install the Stevin app. - Accounts for the business tools you want Stevin to work in, such as GitHub, Linear or Notion. You can connect these later. ## After you are set up - Looking for ideas: see [Common uses](use-cases.md). - Want Stevin to work on a schedule or when something happens: see [Automations](../features/automations.md). - Want to control what it can do on its own and what must be asked first: see [Approvals](../features/approvals.md) and [Permissions](../customize/permissions.md). - Setting up a workspace for a whole team: follow the [setup checklist](../customize/index.md#setup-checklist-for-a-new-workspace). --- path: manual/get-started/what-is-stevin.md title: What is Stevin summary: Stevin is an AI colleague: it takes tasks in team chat, acts in your connected tools, answers the phone, and follows the permissions and approvals you set. --- ## In one sentence Stevin is an AI colleague for your team. It lives in your team's chat tools, in a web console and on the phone. You hand it a piece of work; it looks things up and acts in your connected tools, and asks you first before any action that needs a person's sign-off. ## How it differs from an ordinary chatbot | | Ordinary chatbot | Stevin | |---|---|---| | Where you use it | A separate web page or app | The chat tools your team already uses, the console and the phone | | What it can do | Answer questions, write text | Look up data, create records, send messages and produce files in your connected tools | | Who checks the work | You copy and paste the result yourself | Actions that change data go through [approval](../features/approvals.md) according to your rules | | Afterwards | Over when the chat ends | Every task has a record you can review in [Activity](../features/tasks.md) | | Can it run by itself | No | Yes: [automations](../features/automations.md) run on a schedule or on an event | ## What it can do - **Answer and draft**: answer questions, write documents and organize notes from the conversation and the material in your connected tools. - **Act across tools**: read and write in systems such as GitHub, Linear and Notion through [integrations](../customize/integrations.md) and [MCP servers](../customize/mcp-servers.md). - **Background tasks**: longer work is done in the background and you are notified when it finishes. The request and the result stay in [Activity](../features/tasks.md). - **Automations**: run instructions you have written, triggered by a schedule, a change to a table record, a GitHub event or a webhook. - **Voice and phone**: talk to Stevin [live](../channels/live.md) in the browser, or [call it](../channels/phone.md). Recordings, transcripts and summaries go to the [library](../features/library.md). - **Compute and produce files**: [run a short program](../features/code-execution.md) to total, join or chart data it already has, and hand back a spreadsheet or a chart. - **Work the way your team works**: [skills](../features/skills.md) are workflow instructions written for Stevin, so it handles repeated work by your rules. ## What it will not do - **It does not bypass permissions.** Stevin can only reach tools that are connected and authorized. What it may do there is decided by [permissions](../customize/permissions.md) and by the tool review an admin does for each [MCP server](../customize/mcp-servers.md). - **It does not do risky things quietly.** By default it asks every time before writing. It always asks before deleting, and that cannot be changed to automatic. - **It is not a developer platform.** Stevin is a product you use in chat, in the console and on the phone. There is no public API or SDK for building on it. ## Typical scenarios - Every weekday morning, it posts a summary of yesterday's GitHub merges and open items to a channel. - When a new lead is written to a table, it assigns the lead to the right colleague by your rules and reminds them. - After a meeting, it turns the recording into notes and extracts decisions and action items. - Away from your desk, you call it, ask for a report, and tell it to send the result to your chat when it is done. For more examples, see [Common uses](use-cases.md). --- path: manual/get-started/get-a-workspace.md title: Get a workspace and sign in summary: Request a workspace on the website, then sign in to your team's console with an email code or your Google, Slack, Feishu or Lark account. --- ## Request a workspace 1. Open [stevinhq.com/get-started](https://stevinhq.com/get-started). 2. Fill in the request with your work email and complete the steps on the page. 3. The Stevin team reviews the request. When it is approved, the applicant receives an email and becomes the **Owner** of the workspace. Each team has its own workspace. Its data, integrations and settings belong to that workspace only. ## Sign in to the console The console is the web interface of your workspace. To sign in: 1. Open [stevinhq.com/signin](https://stevinhq.com/signin) and enter your workspace address. It is in your approval or invitation email. 2. Enter your work email and choose **Continue**. 3. Enter the 6-digit code sent to that address. A code is valid for 10 minutes; you can ask for a new one after a short wait. Figure: The sign-in page (Chinese interface shown). Image description: The Stevin sign-in page. A card with an email field, a Continue button, and buttons to sign in with Google or with Feishu, above a language and appearance switch. The interface is shown in Chinese. 1. Work email: Enter the address that was approved or invited. 2. Continue: Sends a 6-digit code to that address. The code is valid for 10 minutes. 3. Sign in with an account you already have: Google, and the chat tool your workspace has connected, once that sign-in method is turned on. 4. Language and appearance: Switch the interface language, or choose system, light or dark. ### Other ways to sign in | Method | When it is available | |---|---| | Email code | Always | | Password | After you set one under **Settings → Account**. Changing it signs you out everywhere | | Google | When your Google account's email matches an account in the workspace | | Feishu or Lark | After an admin turns on sign-in on the [Feishu or Lark integration](../channels/feishu-lark.md#signing-in-to-the-console-with-feishu-or-lark). The first sign-in creates your account | | Slack | After you link Slack once under **Settings → Account**. First sign in another way, then link it | ## What the Owner does next 1. Connect your team's chat tool under **Customize → Integrations**. See [Channels](../channels/index.md). 2. Connect the business tools you want Stevin to use. See [Integrations](../customize/integrations.md) and [MCP servers](../customize/mcp-servers.md). 3. Bring in your colleagues on the **People** page. See [People and roles](../customize/people-and-roles.md). 4. Check that the default **Permissions** suit you. See [Permissions](../customize/permissions.md). The console shows a short **Get started** checklist until these are done. ## How members join After an admin invites you, you receive an email. Sign in with the address that was invited: enter it on the sign-in page and use the code sent to it. If the email does not arrive, ask an admin to choose **Resend invitation** on the People page, or request a code on the sign-in page with the invited address. If your workspace has turned on sign-in with Feishu or Lark, you can use that instead: the first sign-in creates your account, and no invitation is needed. ## Sign-in problems See [Cannot sign in](../troubleshooting/index.md#cannot-sign-in). --- path: manual/get-started/first-conversation.md title: Your first conversation summary: How to hand a task to Stevin, describe it clearly, see what it did, and handle its approval requests. --- ## Start a conversation There are three ways. Use whichever you like: - **In your chat tool**: in a channel or group, mention Stevin and say what you need. In a direct message, just say it. - **In the console**: open **Activity**, choose **New task**, and write a title and a goal. - **By voice**: open **Live** in the console and talk, or [call in by phone](../channels/phone.md). Your first message can be simple: > @Stevin summarize this week's discussion in #product into three conclusions and a list of action items. ## Describe the task clearly - **State the goal and the scope**: "only this week's messages", "only merged pull requests". - **Name the system to use**: "create an issue in Linear", "write the result to the weekly report page in Notion". - **Say where to deliver**: "post it in this channel", "send it to me directly", "produce a file". - **Ask if you are not sure**: "Can you access our GitHub right now?" Stevin tells you which tools it is connected to. If a task is very large, split it into a few smaller ones. Each finishes faster and is easier to check. ## While it works - Simple questions are answered right away. - Work that takes a while runs in the background, and you are notified when it is done. You can carry on with other things. - Actions that need your confirmation stop and wait for you. See the next section. ## Handle approval requests Before Stevin changes data outside the conversation (for example, creating a record or sending a message), it explains what it intends to do and waits for you to confirm. - To agree, choose **Approve**. In a voice conversation, say "yes". - To refuse, choose **Reject**, or say "no". A rejected action is not carried out, and Stevin asks again the next time a similar action comes up. For which actions are asked about and which can run automatically, see [Approvals](../features/approvals.md). ## See what it did Every task is recorded under **Activity** in the console, with its goal, status and result. When a result is not what you expected, open the task there. See [Tasks and activity](../features/tasks.md). ## Start over or tidy up a long conversation In chat, a few commands manage the conversation itself, such as `/new` to start fresh. See [Chat commands](../reference/chat-commands.md). ## Next - [Common uses](use-cases.md): examples you can send as they are. - [Automations](../features/automations.md): have the same job done automatically on a schedule. --- path: manual/get-started/use-cases.md title: Common uses summary: Example tasks you can send as written: summarizing, acting across tools, meetings and calls, files and numbers, and scheduled routines. --- ## How to use this page Every example below is something you can send to Stevin as it is. Replace the channel, repository and table names with your own. The tools involved need to be connected first under [Integrations](../customize/integrations.md) or [MCP servers](../customize/mcp-servers.md). ## Summarizing information - "Summarize today's discussion in this channel into conclusions and action items, and put an owner on each action item." - "Summarize the pull requests merged last week, grouped by module, and write a short update for non-technical colleagues." - "Compare these two documents and list where they conflict." ## Acting across tools - "Create a Linear issue from this piece of user feedback, set the priority to high, and post the link back in this thread." - "Add this week's completed items to the weekly report page in Notion." - "Look up this customer's recent contact history in HubSpot and draft a follow-up email. Show it to me first." These tasks include write actions, so Stevin asks for your [approval](../features/approvals.md) first. ## Meetings and calls - "Turn this recording into notes and extract decisions and action items." Send the recording in chat; the transcript and summary land in the [library](../features/library.md). - Talk in [Live](../channels/live.md) and let it work while you speak. Afterwards the transcript and summary are in the library. - From the car: [call Stevin](../channels/phone.md), ask for what you need, and say "run it in the background and send me the result". ## Files and numbers - "Total the hours per project in this sheet and give me a spreadsheet." - "Draw a weekly trend chart from these figures." - "Check the arithmetic in this quote." Stevin does these with [code execution](../features/code-execution.md), and the files it produces are kept in the [library](../features/library.md). ## Scheduled routines These work well as [automations](../features/automations.md): - At 9:00 every weekday, post a summary of yesterday's GitHub activity to a channel. - When a new lead is added to a table, assign it to the right salesperson by region and remind them. - Every Friday, group the week's customer feedback by product area and write a weekly report. You can also just say it in chat: "From now on, do this at 9:00 every weekday morning." Stevin sets up the automation for you. ## Three habits for writing good tasks 1. Hand over one thing at a time. Split large tasks. 2. Give the criteria for judgment calls, for example "high priority = affects paying customers". 3. For work that recurs, write a [skill](../features/skills.md) so you do not have to explain it every time. --- path: manual/channels/index.md title: Channels summary: Where you can talk to Stevin: team chat (Slack, Feishu, Lark, WhatsApp), the console in your browser, live voice and the phone. --- ## Supported channels | Channel | Good for | Who sets it up | Page | |---|---|---|---| | Slack | Teams that work in Slack | Admin | [Slack](slack.md) | | Feishu or Lark | Teams that work in Feishu or Lark | Admin | [Feishu and Lark](feishu-lark.md) | | WhatsApp | Reaching Stevin from a personal phone in a direct chat | Admin connects, members pair | [WhatsApp](whatsapp.md) | | Console | No chat tool needed: create tasks, review results, approve | Nothing to set up, just sign in | [Console](console.md) | | Live | Talking instead of typing, in the browser | Admin enables voice | [Live voice](live.md) | | Phone | Calling Stevin when you are away from your computer | Each member links a phone | [Phone](phone.md) | ## How to choose - Connect the chat tool your team mainly communicates in first. Most teams only need one. - Feishu and Lark are two editions of the same product, so a workspace connects one or the other, not both. - If you have not decided which chat tool to connect, you can try Stevin with the [console](console.md) alone. ## What channels share Whichever channel a request comes from, Stevin uses the same workspace: - The same [integrations](../customize/integrations.md) and [MCP servers](../customize/mcp-servers.md). - The same [skills](../features/skills.md). - The same [permissions](../customize/permissions.md) and [approval](../features/approvals.md) rules. - Every background task is recorded in [Activity](../features/tasks.md) in the console. ## Recognizing who you are Stevin needs to know that you in the chat tool and you in the console are the same person, so it can act with your role and permissions and send results back to you. Most of the time this happens automatically the first time you message it. If the link is missing, see [Link a chat identity](../customize/people-and-roles.md#link-a-chat-identity). --- path: manual/channels/slack.md title: Slack summary: Install the Stevin app in your Slack workspace, invite it to channels, and use it in channels, threads and direct messages. --- ## Before you connect - You need a Slack workspace admin, or a member who is allowed to install apps. - You need a Stevin workspace and a console account that may manage integrations (Owners and Admins by default). See [Get a workspace](../get-started/get-a-workspace.md). ## Connect Slack 1. In the console, open **Customize → Integrations**. 2. Under **Available**, find **Slack** and choose **Connect**. 3. On the Slack authorization page, check which Slack workspace you are installing into and accept the listed permissions. 4. Back in the console, Slack now shows as **Connected**. 5. In each channel where you want Stevin, invite it: type `/invite @Stevin`. ## Use Stevin in Slack - **In a channel**: mention @Stevin and say what you need. - **In a thread**: keep asking in the same thread and Stevin answers with the earlier context of that thread. - **Direct message**: message Stevin directly. No mention needed. Good for things you do not want to post in a channel. ## What it can see Stevin can only read the channels it has been invited to and the direct messages sent to it. It cannot see channels it has not been invited to. ## Where approvals appear Actions that need your confirmation produce an approval request in the conversation that explains what Stevin intends to do. It continues only after someone approves or rejects. The same request also appears in [Activity](../features/tasks.md) in the console under **Needs review**. See [Approvals](../features/approvals.md). ## Sign in to the console with Slack After you link your Slack account once under **Settings → Account** in the console, you can use **Sign in with Slack** on the sign-in page. If Slack sign-in says you must link first, sign in with your email code, link Slack, and use it from then on. ## Disconnect Choose **Disconnect** on the Slack integration under **Customize → Integrations**. After that Stevin no longer receives or sends Slack messages, and [automations](../features/automations.md) that deliver to Slack fail until you reconnect. ## Problems See [Stevin does not reply](../troubleshooting/index.md#stevin-does-not-reply). --- path: manual/channels/feishu-lark.md title: Feishu and Lark summary: Create a Feishu or Lark app from the console in one step, publish it to everyone, grant its permissions, and use Stevin in groups and direct messages. --- ## Before you connect - Feishu and Lark are two editions that occupy the same slot. A workspace can connect only one of them. To switch, disconnect the current one first. - You need a console account that may manage integrations (Owners and Admins by default). - Publishing the app needs approval from your Feishu or Lark organization admin. ## Connect 1. In the console, open **Customize → Integrations** and choose **Feishu** or **Lark**. 2. Pick one way: - **Create the app in one step** (recommended): the console creates the app for you with the permissions it needs. - **I already have an app**: paste that app's App ID and App Secret. 3. With Lark, the page shows a one-time verification link. Open it in a browser, or scan the QR code with Lark on your phone, then sign in and confirm once. The page updates by itself. 4. Choose **Publish to all members**. A newly created app can be used only by its creator; after publishing, everyone in the organization can find Stevin and message it. 5. Wait for your organization admin to approve the release. Until then only you can use the bot. After approval, return to the console and choose **Check again**. The status changes to **Connected · online**. ## Grant permissions After connecting, the integration page lists each capability Stevin can use in Feishu or Lark, marked as granted or still to grant. Each link opens the Feishu or Lark admin console; an approved permission takes effect immediately. You can also grant everything that is still missing in one click. | Capability | What it lets Stevin do | |---|---| | Docs | Write a document, find one by keyword, and share it with you as soon as it exists | | Base | Build a Base table and add, read, change or remove its records and columns | | Base automation | Start an [automation](../features/automations.md) when a record is added or changed | | Calendar | See your schedule, book a meeting, find free time, reserve a room | | Tasks | Create a task, assign it, follow up and mark it done | | Knowledge base | Write pages into a knowledge base, move documents in, reorganize the tree, and read what is there | | Catch up on messages | Recap what a group discussed while you were away | | Know your teammates | Recognize people by name, so "assign this to Wang" needs no ID | | Command menu | Type `/` in a chat with Stevin to pick a [chat command](../reference/chat-commands.md) | Stevin can only use what has been approved here, and you can revoke access in the Feishu or Lark admin console at any time. ## Use Stevin in Feishu or Lark - **Group chat**: add Stevin to the group, then mention it and say what you need. - **Direct message**: just send a message. No mention needed. - **Commands**: type `/` in the message box to pick conversation commands from a panel. - **Recordings**: send an audio or video recording and ask for notes. The transcript and summary are kept in the [library](../features/library.md). Actions that need your confirmation are sent to you as a card. Use the buttons on the card to approve or reject. ## Signing in to the console with Feishu or Lark An admin can turn on **Sign in with Feishu** (or **Sign in with Lark**) on the integration page. Once it is on, everyone in the organization can sign in to the console with their Feishu or Lark account, and an account is created the first time they do. ## Link your identity If the console says your account is not linked to a chat identity yet, do either of these: - Send Stevin a message in Feishu or Lark. The link is made automatically. - Find yourself on the **People** page in the console and choose **This is me** in the row menu. Only after linking can Stevin reach you on its own initiative, for example to send the result of an automation or of a phone call. ## Disconnect or switch After disconnecting, Stevin can no longer use Feishu or Lark, and [automations](../features/automations.md) that depend on it fail. When you switch from Feishu to Lark (or back), conversations for the whole workspace are received through the newly connected one. --- path: manual/channels/whatsapp.md title: WhatsApp summary: Reach Stevin in a WhatsApp direct chat: an admin links a WhatsApp number, and each member pairs their own phone with a one-time code. --- ## How it works Your workspace links one WhatsApp number of its own. Anyone who messages that number talks to Stevin in a direct chat. Group chats are not supported on WhatsApp. To act with your role and permissions, Stevin has to know which workspace member a WhatsApp sender is. That is what pairing is for. ## Connect the number (admin) 1. In the console, open **Customize → Integrations** and choose **WhatsApp**. 2. Scan the QR code shown there with the WhatsApp app on the phone whose number you want Stevin to use. No WhatsApp business account or developer setup is needed. 3. The integration shows as connected once the phone confirms. ## Pair your phone (each member) 1. In the console, open **Settings → Account** and find **WhatsApp**. 2. Choose **Generate pairing code**. You get a 6-digit code that is valid for 10 minutes and can be used once. 3. From your own WhatsApp, send that code to the workspace's WhatsApp number. From then on Stevin recognizes messages from your phone as you. ## Pair a member's phone for them (admin) On the **People** page, open the member's row menu and choose **Send WhatsApp code**. The code is sent to that member in your team chat tool, and only they can use it. If the member cannot be reached there, the code goes to the workspace Owner to pass on. To remove a wrong number, choose **Unlink WhatsApp** in the same menu. ## Use Stevin in WhatsApp Send a message as you would to a colleague. Background work, approvals and results follow the same rules as in every other [channel](index.md). ## Who can do what Connecting the number needs permission to manage integrations. Sending a member their pairing code and unlinking a number is a separate permission, held by Admins and the Owner by default. See [Permissions](../customize/permissions.md). --- path: manual/channels/console.md title: Console summary: Use Stevin from your browser without any chat tool: create tasks, follow their progress, read results and handle approvals. --- ## When to use it - Your team has not connected a chat tool yet and wants to try Stevin first. - You want to see and manage all tasks in one place. - You need to approve or reject something that is waiting. For signing in, see [Get a workspace and sign in](../get-started/get-a-workspace.md). For everything admins configure in the console, see [Customize](../customize/index.md). ## The pages you use every day | Page | What it is for | Manual | |---|---|---| | **Activity** | Create tasks, follow them, read results, handle approvals | [Tasks and activity](../features/tasks.md) | | **Live** | Talk to Stevin by voice | [Live voice](live.md) | | **Automations** | Create and manage automations | [Automations](../features/automations.md) | | **Library** | Call transcripts, summaries and generated files | [Library](../features/library.md) | Which pages you can see depends on your role and permissions. ## Create a task 1. Open **Activity**. 2. Choose **New task**. 3. Fill in a **Title** (a short name) and a **Goal**: describe exactly what Stevin should do and what result you expect. For example: "Summarize the open GitHub issues and send them to the project lead." 4. Choose **Run task**. The task runs in the background. Changes outside your workspace still wait for approval. For statuses and results, see [Tasks and activity](../features/tasks.md). ## Language and appearance Open the account menu at the bottom left to switch the interface language or choose light, dark or system appearance. See [Your settings](../customize/settings.md). ## On a phone The console works in a mobile browser. For voice, the **Live** page shows a QR code that opens Live on your phone. See [Live voice](live.md#use-it-on-your-phone). --- path: manual/channels/live.md title: Live voice summary: Talk to Stevin in real time from your browser: it listens, works in your tools while you speak, and answers by voice. Sessions are transcribed and recorded. --- ## What Live is Live is a voice conversation with Stevin in the browser. You speak, it listens, does the work and answers by voice. You can ask for lookups and actions across your team's tools, or just talk something through. Every session is transcribed and recorded, and kept in the [library](../features/library.md). ## Start a session 1. Open **Live** in the console and start the session. 2. The browser asks to use your microphone. Allow it. 3. Stevin greets you first. Then just talk. 4. To interrupt while Stevin is speaking, simply start talking. 5. Say "hang up" when you are done, or end the session on the page. Figure: Live, before you start (Chinese interface shown). Image description: The Live page before a session starts. A round blue Stevin avatar in the middle with a Ready label, a short hint and a Start button under it, and Voice and Transcript tabs at the top. The interface is shown in Chinese. 1. Live in the sidebar: Live is its own page in the console, next to Activity. 2. Voice and Transcript: Switch between the voice view and the written transcript of the session. 3. Ready: The state of the session. It changes as Stevin listens, works and speaks. 4. Start: Opens the microphone, after you allow it, and connects. Stevin speaks first. ## Work you hand over during a session Everything you ask for during a session is listed on the page under **Tasks on this session**, marked as a lookup or an action, with its state: | State | Meaning | |---|---| | Working | Stevin is on it | | Needs your yes | Waiting for you to confirm. Nothing is done before that | | Awaiting approval | Paused for approval in **Activity** in the console | | Done | Finished | | Failed | Did not go through. Nothing was done | | Dropped | You did not confirm, so nothing was done | | After the call | Still running when the session ended. The result goes to your chat direct message if your identity is linked | - For items waiting on you, say "yes" or "no", or use **Approve** and **Reject** on the page. - For long jobs, say "run it in the background and send me the result". Stevin keeps working after you hang up and sends the result to your chat. ## Sound quality and devices - **Mute** stops Stevin from hearing you for a moment. - **Hi-fi** switches the microphone to full-band sound with echo cancellation off. Use it with headphones. On speakers, leave it off, or Stevin hears itself. - Under **Settings → Voice** you choose the microphone, whether Hi-fi is on by default, and whether Stevin's voice is lowered the moment you speak. See [Your settings](../customize/settings.md#voice). ## Use it on your phone The Live page shows a QR code under **Open on phone**. Scan it with your phone camera to open Live in the phone's browser. Sign in the first time. Add the page to your home screen to open it full screen like an app. To use a normal phone call instead, see [Phone](phone.md). ## Problems | What you see | What to do | |---|---| | "Voice is not configured for this workspace yet" | Ask an admin to enable voice | | Microphone access was refused, or no microphone is available | Allow the microphone in the browser's site settings and check that a device is connected | | The session was refused | Sign in again and retry | | The connection dropped | Check your network and start again | --- path: manual/channels/phone.md title: Phone summary: Call Stevin from your own phone when you are away from your computer: link your number once, verify each call with an SMS code, then just talk. --- ## What it is for Away from your computer? Call Stevin directly. On a call it can do everything it does in [Live](live.md): answer questions, look things up in your connected tools and carry out actions under the same [approval](../features/approvals.md) rules. Only linked numbers can call in. ## Link your phone (once) 1. In the console, open **Settings** and find **Phone**. 2. Enter the number you will call from, in international format with the country code. 3. Choose **Send verification SMS** and enter the code you receive. It is valid for ten minutes. A number can be linked to one account only. To move it, unlink it from the other account first. To remove it, choose **Unlink phone**. ## Call 1. Call your workspace's Stevin number from your linked phone. The number is shown on the **Live** page under **Call by phone**. 2. Enter the SMS code you receive on the keypad. 3. Once verified, just talk. Say "hang up" when you are done. ## Long jobs while you are on the move You do not have to wait on the line. Say "run it in the background and send me the result". Stevin keeps working after you hang up and sends the result to your direct message in your team's chat tool. For that, your chat identity must be linked; see [Link a chat identity](../customize/people-and-roles.md#link-a-chat-identity). ## Approvals on a call When an action needs confirmation, Stevin asks you on the call and does nothing until you say yes. Actions that need an admin's decision wait in **Activity** in the console. ## After the call The call is recorded and transcribed like a Live session. Find the transcript and summary in the [library](../features/library.md). Who may read other people's transcripts is set in [Permissions](../customize/permissions.md). ## Problems | What you see | What to do | |---|---| | The code is not accepted | Check the SMS and try again. A verification already in progress expires after ten minutes | | "This number is already linked to an account" | Unlink it on the other account first | | The number is rejected | Enter it in international format. Only supported country codes are accepted | | No more verification codes today | The team's daily budget for verification SMS is used up. Try again tomorrow | --- path: manual/features/index.md title: Features summary: What Stevin does day to day: tasks, approvals, automations, skills, memory, the library and code execution. --- ## What this section covers These pages explain how Stevin works from a user's point of view: what each feature is, how to use it, and what to expect. Setting these features up for a workspace (connecting tools, reviewing them, managing people and permissions) is in [Customize](../customize/index.md). ## How the features fit together 1. You hand Stevin a [task](tasks.md), in chat, in the console or by voice. 2. It works with the tools your workspace has connected, following any [skill](skills.md) that applies. 3. Before it changes anything outside the conversation, it asks for [approval](approvals.md). 4. Results come back to you, and files and transcripts are kept in the [library](library.md). 5. Work that repeats becomes an [automation](automations.md). ## Find a feature by what you want to do | I want to | Read | |---|---| | See what Stevin is doing or has done | [Tasks and activity](tasks.md) | | Understand why it asked me before acting, or stop it asking | [Approvals](approvals.md) | | Have something done every day or when an event happens | [Automations](automations.md) | | Teach it how my team does a recurring job | [Skills](skills.md) | | Know what it remembers | [Memory](memory.md) | | Find a transcript or a file it made | [Library](library.md) | | Get numbers totalled, data joined or a chart drawn | [Code execution](code-execution.md) | --- path: manual/features/tasks.md title: Tasks and activity summary: Everything Stevin works on in the background is a task. The Activity page shows each task's status and result, and is where you handle approvals. --- ## What a task is One piece of work Stevin does in the background is one task. A task can come from three places, and all of them land on the same queue: | Origin | How it starts | |---|---| | Chat | You ask in a chat tool, in Live or on the phone, and the work takes more than a quick reply | | Console | You choose **New task** on the Activity page | | Automation | A trigger of an [automation](automations.md) fires | Simple questions are answered on the spot and do not become tasks. Work that takes several steps runs in the background; you do not have to wait, and you are notified when it is done. ## Task statuses | Status | Meaning | What you need to do | |---|---|---| | Queued | Waiting to start | Nothing | | Running | Stevin is working on it | Nothing | | Awaiting approval | An action is waiting for a person to confirm | Approve or reject | | Done | Finished; the result is available | Read the result | | Failed | Went wrong partway and did not finish | Read the error, fix the cause and run it again | | Cancelled | Was cancelled or rejected and did not continue | Nothing | ## The Activity page Open **Activity** in the console to see the tasks you are allowed to see, newest first, grouped by day. - The tabs filter the list: **All**, **Automations**, **Waiting on you**, **Failed** and **Done**. - When new tasks arrive while you are reading, the page offers to show the latest instead of moving the rows under you. Open a task to see: - **Result**: the final reply or output. - **Goal**: what the task was asked to do. - Its status, where it came from, when it was created and finished, how long it took, and the model used. - For a task started by an automation, a link to that automation. ## Create a task in the console 1. Choose **New task**. 2. Enter a **Title** and a **Goal**. In the goal, describe exactly what Stevin should do and what result you expect. 3. Choose **Run task**. The task runs in the background. Changes outside your workspace still require approval. ## Tasks that need review A task that wants to make a change outside your workspace waits under **Needs review**. Open it to read what Stevin intends to do, then: - **Approve**: Stevin carries out the action and continues. - **Approve & always allow**: approves this one and lets the same kind of action run without asking from now on. - **Reject**: the action is not carried out and the task is cancelled. The same request appears in the chat conversation where the task started, so you can decide in either place. Some actions must be decided by an admin other than the person who asked; in that case you see "Another admin must approve this action; you may reject it." For the rules, see [Approvals](approvals.md). ## When a task does not finish - **The outside system refused**: the connected account lacks access there. Check the authorization of the [integration](../customize/integrations.md) or [MCP server](../customize/mcp-servers.md) involved. - **A tool is missing or switched off**: the tool was never allowed, or came back for review after the provider changed it. See [MCP servers](../customize/mcp-servers.md#when-a-provider-changes-its-tools). - **The allowance is used up**: when the workspace's credits for the period run out, Stevin pauses until the allowance resets or the plan is upgraded. See [Usage and plans](../customize/usage-and-plans.md). - **The task was too large**: split it into smaller tasks and run them one by one. After fixing the cause, create the task again with the same goal. ## Who can see and manage tasks By default, Members, Admins and the Owner see everyone's tasks and Viewers see only their own. Members can create, cancel and approve their own tasks; Admins and the Owner can do so for everyone's. All of this can be changed on the [Permissions](../customize/permissions.md) page. --- path: manual/features/approvals.md title: Approvals summary: Stevin asks before it changes anything: which actions need approval, where you approve, who can approve, and how to let a trusted action run automatically. --- ## Why approvals exist Stevin can really act in your tools, so risky actions need a person to check them. Approvals guarantee one thing: it does not change your data without consent. ## Three kinds of action Every tool Stevin can use is classed by its effect, and the rules differ: | Class | Examples | Default rule | Can it be made automatic | |---|---|---|---| | Read | Look up an issue, read a document | Runs directly | Not applicable | | Write | Create a record, send a message, change a field | Asks first | Yes | | Delete and irreversible | Delete a record, delete a file | Always asks first | No | A delete cannot be undone, so it always pauses for approval, and nobody can change that. ## What an approval request looks like Stevin stops and explains what it intends to do: which tool, on what, with what content. It acts only after a person confirms. You can decide in any of these places: - **In your chat tool**: on the approval card in the conversation. - **In the console**: the task waits under **Needs review** on the **Activity** page. - **In a voice conversation**: say "yes" or "no", or use the buttons on the Live page. ## After you decide - **Approve**: Stevin carries out the action and continues with the following steps. - **Approve & always allow**: approves it and lets this kind of action run without asking in future. - **Reject**: the action is not carried out and nothing is changed. A rejection is not remembered as a rule: the next time a similar action comes up, Stevin asks again. A request that nobody handles keeps waiting. It is never carried out automatically. ## Who can approve - The person who started a task can approve the actions in their own task, as long as their role is allowed to approve tasks. - Some actions are escalated beyond the person who asked. An escalated action must be decided by an Admin or the Owner who is not the person who asked. That person can still reject it. - By default, Members approve their own tasks and Admins and the Owner approve anyone's. Viewers cannot approve. Change this with the **Approve what a task is waiting to do** permission on the [Permissions](../customize/permissions.md) page. ## Being asked less often By default every write is asked about. For tools and actions you already trust, there are three ways to cut the questions: - **Per action, as you go**: choose **Approve & always allow** on a request. - **Per tool**: on an [MCP server](../customize/mcp-servers.md#review-the-tools), an admin sets a write tool to **Runs directly** instead of **Asks first**. - **Per role, for conversation-driven setup**: on the [Permissions](../customize/permissions.md#what-stevin-may-decide-in-a-conversation) page, an admin chooses how often Stevin asks before it installs an MCP server or switches on one of its tools from a conversation: ask every time, ask for anything new, or never ask. Start with asking every time and loosen gradually once you trust its judgment for a given action. ## Approvals in automations When an [automation](automations.md) reaches an action that needs approval, it also stops and waits. An automation meant to run unattended should either use only read tools or have the write tools it needs set to run directly beforehand. ## The record Every approval request and its outcome is part of the task it belongs to, and can be reviewed in [Activity](tasks.md). --- path: manual/features/automations.md title: Automations summary: Have Stevin run an instruction on a schedule or when something happens: triggers, what one run does, which tools it may use, limits, and run history. --- ## What an automation is An automation is a trigger plus one instruction. At the set time, or when the event happens, Stevin follows the instruction you wrote and delivers the result. It suits work that repeats, such as a morning briefing, a weekly report, assigning new leads, or checks after a release. ## Create an automation There are two ways: - **Say it in chat**: "From now on, at 9:00 every weekday morning, summarize the pull requests merged yesterday and post them in this channel." Stevin sets it up for you. - **Build it in the console**: open **Automations**, create a new one, and fill it in as described below. ## Triggers An automation can have several triggers. It runs when any one of them is met. | Trigger | When it fires | Setup | |---|---|---| | Schedule | Every day, every weekday, chosen days of the week, or monthly, at the time you set | Pick it in the visual editor, or write a recurrence rule (RRULE) for anything more complex. Times are in your browser's time zone | | Feishu or Lark Base record | A record is added to the chosen Base table, or changed | Needs the Base automation capability on the [Feishu or Lark integration](../channels/feishu-lark.md#grant-permissions). The Base-side workflow is created and removed for you | | GitHub event | A selected event happens in the chosen repository, such as a new pull request or issue | Uses your connected [GitHub integration](../customize/integrations.md#github). Nothing to paste on GitHub's side | | Webhook | Another system calls the automation's own address | See the next section | ## Webhook triggers When you save an automation with a webhook trigger, it gets its own address and key. Both appear on the automation's page. - Send the key in an `Authorization: Bearer` header or an `X-Webhook-Secret` header. A header is the safer choice because it stays out of logs and browser history. - For systems whose whole configuration is one URL box, the page also gives an address with the key already in it. - **Filter**: optionally, a regular expression matched against the JSON body of each delivery. Only deliveries that match start a run, so a busy source can be narrowed to the events you want. Leave it empty to run on every delivery. - **Rotate the key** at any time. The address stays the same; give the new key to every sender. Treat the key as a password: anyone who has it can make the automation run. ## The instruction Write what **one run** should do, in plain language, as you would brief a colleague. How often it runs comes from the triggers. Make clear: - What to read. - How to process it and what to compare or judge by. - Where to send the result and in what format. ## Run settings | Setting | What it does | |---|---| | MCP servers | Which installed [MCP servers](../customize/mcp-servers.md) this automation may use. Select only what it needs. Workspace permissions still apply | | Model | The model for future runs. By default the workspace's automation model; see [Default models](../customize/models.md) | | Run limit | Caps how many times it may run in a period. A schedule can run uncapped; event-driven automations get a default ceiling so an event storm cannot run away | ## How runs behave - Each run is a [task](tasks.md) and appears in **Activity** under the **Automations** filter. - Actions that need approval still stop and wait for someone to confirm. For an automation to run unattended, the write tools it uses must be set to run directly. See [Approvals](approvals.md#approvals-in-automations). - Runs over the limit are skipped. ## Test runs and history - **Run now** starts a run immediately with the saved instruction. For an event-driven automation you can add context that the run should take into account. - The automation's page shows its recent runs with their outcome, and a chart of the last days. ## Pause, change or cancel - You can edit the instruction, triggers and settings at any time; future runs use the new version. - Cancelling an automation stops it and takes down any webhook it created. - When an integration it depends on is disconnected, its runs fail. They recover once you reconnect. ## Advice 1. Do the job once by hand in chat first, and turn it into an automation once you are happy with the result. 2. Write "do not send a message when there is nothing new" in the instruction, to avoid empty messages. 3. Set a run limit for frequent events. --- path: manual/features/skills.md title: Skills summary: Skills are instructions you write for Stevin so it does recurring work the way your team does. How skills work and how to write a good one. --- ## What a skill is A skill is a set of instructions that tells Stevin how to do a kind of job: when it applies, the steps, what to watch for, and what the deliverable looks like. When Stevin recognizes a matching task, it reads the skill before it starts. Think of it as the operating manual you would give a new colleague. Write it once and it is followed every time, with no repeated explaining. ## When to write a skill - You have explained the same kind of task to Stevin three times or more. - Your team has a fixed format or rule for it, such as a weekly report template or a ticket severity standard. - Mistakes are costly and it must follow the steps strictly, as with refunds. ## Team skills and your own skills | Type | Who writes it | Who it applies to | |---|---|---| | Your own skill | Any member | Only you: only your conversations use it | | Team skill | Admins | Everyone in the workspace | Once a skill of your own has proved useful, an admin can publish it to the team. Creating, switching on and publishing skills is done in the console; see [Manage skills](../customize/skills.md). ## The parts of a skill | Part | What to write | |---|---| | Name | Short, lowercase, with hyphens, such as `weekly-report` | | What it does | One or two sentences on the situations it applies to. Stevin uses this to decide when to use the skill, so be specific | | Full instructions | What Stevin reads when it uses the skill | | Extra material | Optional files with finer detail (templates, examples), which Stevin opens only when it needs them | ## A template you can edit ```markdown ## When to use When someone asks for a weekly report or mentions "this week's summary". ## Steps 1. Confirm the time range; the default is this Monday to today. 2. Read merged pull requests from GitHub and completed issues from Linear. 3. Organize them into three parts: Done / In progress / Risks. 4. Post to the place the person named; if none, reply in the current conversation. ## Notes - One line per item at most, with a link. - Write "None" for an empty part; do not leave it out. ``` ## What makes a good skill - **The description decides whether it gets used.** Spell out the triggering situation rather than a vague "handles weekly reports". - **Steps must be executable.** Write "read field Y of table X", not "understand the situation". - **State the limits.** Say when to stop and ask a person, and what not to do. - **One skill, one job.** Split one that grows too long. ## Skills and permissions A skill is only instructions. It gives Stevin no extra access. The tools a skill uses still have to be connected and reviewed, and the write actions in it still go through [approval](approvals.md). --- path: manual/features/memory.md title: Memory summary: What Stevin remembers within a conversation, how to start fresh or shorten a long one, and how to make it keep your team's rules. --- ## Within a conversation Within one conversation, Stevin remembers what was said earlier. You can say "do another one in the same format" or "change the second item". In a long conversation, earlier content is condensed into a summary that keeps the main points. If it has lost an earlier detail, say it again. ## Manage a conversation yourself In chat, a few [commands](../reference/chat-commands.md) act on the conversation itself: | Command | Use it to | |---|---| | `/new` | Start a fresh conversation. The old one is archived, not deleted | | `/compact` | Fold the earlier turns into a summary and carry on. Add a focus to keep, for example `/compact keep the pricing` | | `/context` | See what this conversation's context currently holds | | `/undo` | Bring back the conversation that the last `/new` archived. Send it again to swap back | A command only affects the conversation it is sent in: your own direct-message conversation, or the group thread you sent it in. It never touches anyone else's conversation, and files are untouched. Start a new conversation when you change topic. Stevin answers better, and faster, when the conversation holds only what matters for the task. ## Across conversations A new conversation does not carry over what was said in another one. For anything Stevin should know every time, do not rely on it remembering a chat: write it down where it always looks. ## Make it keep your team's rules The reliable way to make Stevin follow a rule every time is a [skill](skills.md): - Your team's fixed formats and templates. - Definitions of terms and internal names. - The criteria for handling a kind of case. Skills are plain text with a revision history, so anyone can see what Stevin is working from, and changes are easy to follow. ## What it cannot recall - Content in systems that are not connected. - Content the person asking is not allowed to see. Whose data Stevin may read on someone's behalf is set under [Permissions](../customize/permissions.md). For how data is handled, see [Security and data](../reference/security-and-data.md). --- path: manual/features/library.md title: Library summary: Transcripts and summaries of your calls and recordings, and the files Stevin generates, are kept in the Library, where you can open, filter and download them. --- ## What is in the library The **Library** page in the console keeps two kinds of content in one place: - **Transcripts**: the record of [Live](../channels/live.md) sessions, [phone](../channels/phone.md) calls, and recordings you send to Stevin in chat. - **Files**: what Stevin produced in tasks, such as reports, spreadsheets and charts. ## Transcripts Each session or recording is saved with: | Content | Description | |---|---| | Transcript | The full text of what was said | | Summary | The main points | To turn a meeting recording into notes, send the audio or video to Stevin in chat and say what you want from it. The transcript and summary land in the library. ## Files Files generated in a task are attached to Stevin's reply and also kept in the library, where you can open or download them later. To get a file, say in the task what form you want: - "Put the result in a spreadsheet file." - "Draw a weekly trend chart." - "Write it as a document I can send straight to the customer." When data needs to be calculated, summarized or charted, Stevin uses [code execution](code-execution.md). ## Find something Use the filters on the Library page to narrow the list, for example by date range. If the library is empty, nothing has been recorded or generated for you yet: send a recording to Stevin or start a Live session. ## Who can see what By default, Viewers and Members see their own transcripts, and Admins and the Owner see everyone's. Publishing a transcript (for example, posting it to a group) and saving files follow the same pattern: Members for their own, Admins and the Owner for everyone's. An admin can change this on the [Permissions](../customize/permissions.md) page. --- path: manual/features/code-execution.md title: Code execution summary: Stevin can write and run a short Python program to compute over data it already has: total, join, rank, check arithmetic, or produce a file. --- ## What it is for Language models are unreliable at arithmetic over many rows. For that kind of work Stevin writes a short Python program and runs it, then reports the result. Typical uses: - Total, average or rank the rows of a table. - Join two lists on a shared field. - Check the arithmetic in a quote or a report. - Produce a file: a spreadsheet, a CSV, a chart. You do not have to ask for code. Ask for the result ("total the hours per project and give me a spreadsheet") and Stevin decides when a program is the right tool. ## What the program can and cannot reach The program runs in a sandbox, separate from your systems: - It works on data Stevin already has in the conversation or has fetched through your connected tools. - It has no network access, so it cannot call outside services or send data anywhere by itself. - Files it produces come back through Stevin, are attached to the reply and are kept in the [library](library.md). ## Installing packages When a program needs a library the sandbox does not have, Stevin can install a Python package into the sandbox. - Installing a package asks for approval, like any other action with an effect. - An admin can list package names that install without approval (for example `pandas`, `openpyxl`), on the **Install packages** integration. - Who may have packages installed, and who may change the list, are two separate [permissions](../customize/permissions.md). ## Turn it on or off **Code execution** and **Install packages** are listed under **Customize → Integrations** in the console. An admin can connect or disconnect each one. See [Integrations](../customize/integrations.md#built-in-capabilities). --- path: manual/customize/index.md title: Customize summary: Everything you set up in the console: what Stevin is connected to, what it has been taught, who is in the workspace, what each role may do, and your plan. --- ## What this section covers The console is where a workspace is shaped. This section gathers every control-panel operation in one place, so you can find how to change a setting without knowing which feature it belongs to. | I want to | Page | Where in the console | |---|---|---| | Connect a chat tool, GitHub or a built-in capability | [Integrations](integrations.md) | **Customize → Integrations** | | Add an outside service and decide which of its tools Stevin may use | [MCP servers](mcp-servers.md) | **Customize → MCP** | | Create, switch on, publish or roll back skills | [Manage skills](skills.md) | **Customize → Skills** | | Choose which model answers chat, tasks and automations | [Default models](models.md) | **Settings → Default models** | | Invite people, set roles, link chat identities | [People and roles](people-and-roles.md) | **People** | | Decide what each role may do | [Permissions](permissions.md) | **Permissions** | | See usage, change plan, manage billing | [Usage and plans](usage-and-plans.md) | **Settings → Plan and usage** | | Change your own password, language, appearance, microphone, phone | [Your settings](settings.md) | **Settings** and the account menu | ## Find your way around the console The sidebar has the pages everyone uses (**Activity**, **Live**, **Automations**, **Library**) followed by the ones that shape the workspace (**Customize**, **People**, **Permissions**). The **Customize** page has three tabs, **Integrations**, **Skills** and **MCP**, each split into what is connected and what is available. Your account menu and **Settings** are at the bottom left. Which pages and controls you see depends on your role. If something described here is missing for you, your role does not include it; see [Permissions](permissions.md). ## Who can change what | Area | Owner | Admin | Member | Viewer | |---|---|---|---|---| | Integrations, MCP servers, tool review | Yes | Yes | No | No | | Team skills | Yes | Yes | Read | No | | Their own skills | Yes | Yes | Yes | No | | Default models, usage and billing | Yes | Yes | No | No | | See people in the workspace | Yes | Yes | Yes | Own row | | Change roles, enable or disable accounts | Yes | No | No | No | | Edit the Permissions page | Yes | No | No | No | These are the defaults. Except for the Owner's, they can be changed on the Permissions page. ## Setup checklist for a new workspace 1. Connect a chat tool: [Channels](../channels/index.md). 2. Connect business tools: [Integrations](integrations.md) and [MCP servers](mcp-servers.md), and review the tools of each MCP server. 3. Bring in your team and assign roles: [People and roles](people-and-roles.md). 4. Check the default permissions: [Permissions](permissions.md). 5. Write your first team skill: [Manage skills](skills.md). 6. Try building an automation: [Automations](../features/automations.md). ## Regular checks Once a month, look at: - Whether any integration or MCP server is no longer used and can be removed. - Whether any write tool set to run directly no longer needs to be. - Whether any team skill describes an outdated process. - Whether people who left still have accounts. --- path: manual/customize/integrations.md title: Integrations summary: Connect your team's chat tool, GitHub and Stevin's built-in capabilities on the Customize page: what is available, how to connect, and how to disconnect. --- ## What integrations are Integrations decide which systems Stevin can see and act in. A system that is not connected can be neither read nor changed. There are two places to connect things, both on the **Customize** page: - The **Integrations** tab: chat tools, GitHub and a few built-in capabilities. This page. - The **MCP** tab: any other outside service, added as an MCP server. See [MCP servers](mcp-servers.md). ## What you can connect here | Integration | What it gives Stevin | Details | |---|---|---| | Slack | Receive requests and send results in your Slack workspace | [Slack](../channels/slack.md) | | Feishu or Lark | Receive requests and send results in Feishu or Lark, plus Docs, Base, Calendar, Tasks and the knowledge base. Only one of the two can be connected | [Feishu and Lark](../channels/feishu-lark.md) | | WhatsApp | A direct chat with Stevin on your workspace's own WhatsApp number | [WhatsApp](../channels/whatsapp.md) | | GitHub | Read repositories and work with issues and pull requests | [GitHub](#github) | | Code execution | Write and run a short Python program over data it already has | [Code execution](../features/code-execution.md) | | Install packages | Install Python packages into the code sandbox when a program needs one | [Code execution](../features/code-execution.md#installing-packages) | The complete and current list is the one shown in your console. ## Connect an integration 1. Open **Customize** in the console and stay on the **Integrations** tab. It is divided into **Connected** and **Available**. 2. Under **Available**, choose the integration and choose **Connect**. 3. Complete authorization as the page directs. Some integrations send you to the provider's own authorization page; others ask for app credentials. Secrets you enter are stored on the server and are not shown again. 4. The integration moves to **Connected**. Open it to see its status and settings. Connecting, disconnecting and configuring integrations needs the matching permission, held by Admins and the Owner by default. Figure: Customize → Integrations (Chinese interface shown). Image description: The Customize page on its Integrations tab. A table lists four connected integrations (Feishu, GitHub, Code execution, Install packages), each with a status. The interface is shown in Chinese. 1. Customize: The page that holds integrations, skills and MCP servers. 2. Integrations tab: The first of three tabs. Skills and MCP are next to it. 3. Connected and Available: Switch between what is already connected and what you can add. The numbers are counts. 4. A connected integration: Its name and a one-line description. Select the row to open its settings. 5. Status: Connected, or what still needs attention. A chat tool also shows whether it is online. 6. More actions: Open, disconnect and other actions for this integration. ## GitHub With GitHub connected, Stevin can read repositories and work on issues and pull requests. - **Default repository**: when a conversation does not name a repository, Stevin uses this one. - **Automations**: a connected GitHub account is what lets an [automation](../features/automations.md#triggers) start on a GitHub event. The hook is created for you. ## Built-in capabilities Besides outside tools, the Integrations tab has capabilities that come with Stevin: - **Code execution**: lets Stevin write a short program to total, join or rank data it already has, or to produce a file. The program has no network access. - **Install packages**: installs a Python library into the sandbox when a program needs one. Admins can list the packages that install without approval. See [Code execution](../features/code-execution.md). ## Least privilege - Connect only the tools you really need. - When authorizing, grant only the scope that is needed, for example the relevant repositories rather than the whole organization. - For each [automation](../features/automations.md#run-settings), select only the MCP servers it needs. ## Disconnect an integration Open the integration under **Connected** and choose **Disconnect**. After that: - Stevin immediately loses every capability of that integration. - Automations that depend on it fail, and recover once you reconnect. ## When authorization goes wrong When Stevin replies that it has no permission, or the outside system refuses, open the integration and check its status. Changes on the other side (for example, an admin revoking the app's permissions) also break it; authorize again to fix it. If the page says statuses could not be checked, that alone does not mean the connections are down: reload and look again. --- path: manual/customize/mcp-servers.md title: MCP servers summary: Add outside services as MCP servers: install one, connect an account, then review its tools one by one and decide which run directly and which ask first. --- ## What an MCP server is MCP (Model Context Protocol) is an open standard for making a service usable by AI. One MCP server corresponds to one service, such as Linear or Notion, and offers a number of tools, such as "look up an issue" or "create an issue". Remember one thing: **installed does not mean usable. Stevin can only call tools that have been allowed in the review.** ## Two sources On **Customize → MCP**: - **Managed**: servers Stevin manages for you, with a recommended set of tools. Pick one and install it. - **Custom MCP**: any remote MCP server reachable over HTTPS, added by its address. Use it for a vendor's official server or one your organization hosts. ## From install to usable 1. **Install.** Choose a managed server and choose **Install**, or register a custom one (see below). Installing only records the choice: nothing is connected and no tool is permitted yet. 2. **Connect.** Choose **Connect** and complete authorization, so that Stevin reaches the service as your organization's account. **Test connection** checks that it works. 3. **Review the tools.** Decide, tool by tool, what Stevin may call. | Status | Meaning | |---|---| | Not installed | Listed, not added to your workspace | | Awaiting authorization | Installed; no account connected yet | | Awaiting tool review | Connected; tools not reviewed yet | | Available | Ready to use | | Needs review | Its tools changed and must be reviewed again | | Connection failed | Stevin cannot reach it; reconnect or test the connection | | Disabled | Switched off by an admin; settings are kept | ## Add a custom MCP server On the **Custom MCP** tab, register the server: | Field | What to enter | |---|---| | Identifier | Lowercase letters, digits and hyphens, at most 24 characters. It prefixes every tool name Stevin sees | | Endpoint URL | The server's HTTPS address | | Description | Plain words on what this server is for, such as "Company GitHub, the repositories under the acme org". It helps Stevin tell apart several servers of the same kind | | Authentication | **Automatic** by default: the server's own answer decides between no credential, a Bearer token (API key) and standard MCP OAuth, where you sign in on the vendor's page. Extra request headers can be added when connecting | Then choose **Register and install**. An address you registered but have not installed stays on the Custom tab until you install it or choose **Forget**. ## Review the tools After connecting, the server's page lists its tools in three groups: | Group | Who can trigger them | Before running | |---|---|---| | Read | Members whose role has read access | Runs directly once allowed | | Write | Members whose role has write access | **Asks first** by default; can be set to **Runs directly** | | Delete and irreversible | Members whose role has delete access | Always pauses for approval; cannot be changed | - Switch each tool on or off under **Allowed**, or use **Allow all** and **Deny all** and then fine-tune. A search box filters long lists. - Every change saves immediately. - Stevin neither sees nor uses a tool that is not allowed. Figure: Reviewing the tools of an MCP server (Chinese interface shown). Image description: The page of one MCP server, Linear, showing that it is connected with 70 tools. Below, the Read-only group lists tools such as Get issue and Get document, each with a switch. The interface is shown in Chinese. 1. Connection status: Whether the server is connected, and how many tools it offers. 2. Enable or disable: Switches the whole server off without losing its settings. 3. All, Allowed, Not allowed: Filter the tools by their review state. 4. Read-only group: Tools that only read. The count shows how many are allowed. Write and delete tools have their own groups below. 5. Allow all: Allow or deny the whole group at once, then fine-tune. 6. One tool: Its name, what it does, and its switch. Stevin cannot see a tool that is switched off. Deciding which tools Stevin may call is its own permission, separate from installing and connecting. By default Admins and the Owner hold both. ## When a provider changes its tools Stevin can only call a tool with the arguments it had when you reviewed it. A tool that changes, or a new one, comes back for review and is not used until an admin allows it. This prevents a tool nobody has looked at from being used quietly. ## Installing from a conversation A member can ask Stevin in a conversation to install and connect an MCP server, or to switch on a tool of an installed one. Whether it asks for confirmation first is set per role on the [Permissions](permissions.md#what-stevin-may-decide-in-a-conversation) page. These settings only remove a confirmation; who may install or review at all is still decided by the console permissions. ## Disable or uninstall - **Disable** switches a server off without losing its settings. **Enable** brings it back. - **Uninstall** removes it. Stevin loses access to its tools immediately; to use it again, install, connect and review it again. - A server marked **Required by** an integration can only be switched off by an admin. ## Security advice - Add only custom servers you trust. Stevin reads what they return. - For a new server, allow only the read tools at first, and open up writes after watching it for a while. - Look through the installed list from time to time and remove what is no longer used. --- path: manual/customize/skills.md title: Manage skills summary: Create, switch on, publish and roll back skills on the Customize page: team skills, your own skills, and members' skills. --- ## Where skills are managed Open **Customize → Skills** in the console. For what a skill is and how to write a good one, see [Skills](../features/skills.md). The page has up to three lists: | List | What is in it | Who sees it | |---|---|---| | **Team** | Skills every member's conversations use | Everyone who can see skills | | **Mine** | Skills only your own conversations use | You | | **Members' own skills** | Skills members wrote for themselves | Admins only, so they can publish a good one to the team | Figure: Customize → Skills (Chinese interface shown). Image description: The Customize page on its Skills tab. The Team list shows skills such as automations, data-analysis and feishu-calendar, each with a one-line description and an Enabled status. The interface is shown in Chinese. 1. Skills tab: Between Integrations and MCP. 2. Team, Mine, Members' own: The three lists, each with a count. 3. New skill: Opens the form to create a skill. 4. A skill: Its name and what it does. Select the row to open, edit or see revisions. 5. Status: Whether the skill is switched on. ## Create a skill 1. Choose **New skill**. 2. Fill in the form: | Field | What to enter | |---|---| | Name | Lowercase letters, digits and hyphens, such as `weekly-report` | | What it does | The situations the skill applies to. Stevin decides from this when to use it | | Full instructions | What Stevin reads when it uses the skill | | Extra material | Optional files the instructions refer to. Stevin opens them only when it needs them | 3. Choose whether it is a team skill. A skill that is not is yours alone until an admin publishes it to the team. 4. Choose **Save**. ## Switch a skill on or off Each skill has an on/off switch. A skill that is off keeps its content but is not used. Some skills are marked **Required by** an integration or by the product: only an admin can change them or switch them off. ## Publish a member's skill to the team When a member's own skill has proved useful, an admin opens it and chooses **Publish to team**. From then on every member's conversations use it. ## Revisions Every save creates a revision. Open a skill and choose **Revisions** to see every saved version, newest first. Choose **Make active** on an earlier one to go back to it. If the active revision cannot be read, the page says so: make an earlier one active or edit it. ## Delete a skill Choose **Delete** on the skill. Its revisions stay in the audit record. ## Who can manage skills | Action | Owner | Admin | Member | Viewer | |---|---|---|---|---| | See the team's skills and their own | Yes | Yes | Yes | No | | Create and edit their own skills | Yes | Yes | Yes | No | | Create, edit and publish team skills | Yes | Yes | No | No | These are the defaults; change them on the [Permissions](permissions.md) page. --- path: manual/customize/models.md title: Default models summary: Choose which AI model Stevin uses for chat replies, for background tasks and for automations. --- ## Three settings A workspace has one default model for each of three situations. Change them under **Settings → Default models** in the console. | Setting | Used for | Takes effect | |---|---|---| | Chat model | Immediate replies in chat | From the next chat turn | | Background task model | Tasks that run in the background, and actions carried out after an approval | For newly queued tasks | | Automation model | Runs of automations | For future runs | An individual [automation](../features/automations.md#run-settings) can also choose its own model. ## Choosing Models differ in speed, capability and how many credits they use. A stronger model on every automation run is the most common reason credits go faster than expected; check **Usage by model** on the [Usage and plans](usage-and-plans.md) page before and after a change. If unsure, keep the defaults. If a model you chose earlier is no longer offered, the setting shows it as unavailable and the workspace default is used until you pick another. ## Who can change models Switching models is its own permission, held by Admins and the Owner by default. See [Permissions](permissions.md). --- path: manual/customize/people-and-roles.md title: People and roles summary: Bring people into the workspace, assign the four roles (Owner, Admin, Member, Viewer), link chat identities, and disable accounts. --- ## Four roles | Role | Who it is for | What it can do by default | |---|---|---| | Owner | The person responsible for the workspace | Everything. The only role that can change roles and edit permissions. Its permissions are fixed | | Admin | People who configure and manage | Connect integrations, install and review MCP servers, manage team skills, switch models, see usage and billing, sync the directory | | Member | People who use Stevin day to day | Talk to Stevin, create and approve their own tasks, write their own skills, see the team's tasks | | Viewer | People who only need to keep informed | See their own tasks and transcripts. Cannot talk to Stevin or act | These are the defaults. Except for the Owner's, what each role can do can be changed on the [Permissions](permissions.md) page. ## The People page The **People** page lists everyone Stevin can recognize, and shows who may use the console. - **Can sign in**: the person has a console account. - **Directory only**: the person is known from your chat tool's directory but has no console account. Stevin can still recognize them in chat. - Filter the list by **Everyone**, **Can sign in** or **Directory only**. Each row shows the person's role and their linked chat identities. ## Bring people in There are two ways, and you can use both: - **Sync the directory.** With Feishu or Lark connected and the contact permission granted, choose **Sync from contacts** to bring in everyone from your organization's directory. - **Invite by email.** Choose **Invite by email**, enter the person's email address (and optionally their name) and send. They sign in with a code sent to that address. ### Manage an invitation | Action | When to use it | |---|---| | **Resend invitation** | The email did not arrive | | **Change email and re-invite** | The address was wrong and has never been signed in with | If the page says the account was opened but the email did not go out, tell the person to sign in on the sign-in page with that address. ## Change a role In the person's row, choose a new role. The change takes effect immediately. Only the Owner can change roles by default. ## Link a chat identity Stevin needs to know which person in the chat tool a console account corresponds to. Without the link it cannot find that person in chat, for example to mention them or to send them a result. - **Yourself**: send Stevin a message in the chat tool and the link is made automatically. Or find yourself on the People page and choose **This is me** in the row menu. - **Others**: for people shown as not linked, ask them to do either of the above. Your own link is also shown under **Settings → Chat identities**. ## Disable or enable an account Choose **Disable** in the person's row menu. After that: - The person is signed out of the console and cannot sign in again. - The account and its history are kept, and **Enable** restores access at any time. To stop someone using Stevin in your chat tool as well, remove them there too. ## WhatsApp pairing Admins can send a member their WhatsApp pairing code, or unlink a wrong number, from the row menu. See [WhatsApp](../channels/whatsapp.md). ## Advice - Keep Owners and Admins few. - When someone leaves, disable their account, then check the [automations](../features/automations.md) they created. --- path: manual/customize/permissions.md title: Permissions summary: The Permissions page decides what each role may do: console actions, whose data a role can reach, and how often Stevin asks before deciding something in a conversation. --- ## What the page is The **Permissions** page is a table: one row per permission, one column per role (Owner, Admin, Member, Viewer). Each cell says what that role may do. Changing a cell changes it for everyone with that role, immediately. By default only the Owner can edit this page. The Owner column itself is fixed, so a workspace can never lock everyone out. Figure: The permission matrix (Chinese interface shown). Image description: The Permissions page. A table with one row per permission, grouped under a Console heading, and one column each for Owner, Admin, Member and Viewer with a checkbox in every cell. The interface is shown in Chinese. 1. All and Changed: Show every permission, or only the ones that differ from the default. 2. Recent changes: Who changed which permission, when, and what it was before. 3. Role columns: One column per role, with the number of people in it. The Owner column is locked. 4. Group: Permissions are grouped. The counts on the right show how many of the group each role has. 5. Permission: What the row controls, in plain words. The question mark explains it further. 6. A cell: The value for one role. Select it to change it; it is saved at once. ## Three kinds of value | Group | Values | The question it answers | |---|---|---| | Console | **Allow** or **Deny** | May this role do this at all? | | Data and the agent | **no one**, **their own**, **everyone** | Over whose tasks, records, transcripts or skills? | | What Stevin may decide in a conversation | **ask every time**, **ask for anything new**, **never ask** | How often does Stevin ask before doing this from a conversation? | A role higher on the ladder does not automatically get more: each cell is set on its own. An unknown or missing value always means no access. ## Console permissions These decide who can see and change parts of the console: integrations, MCP servers and their tool review, team usage and billing, default models, directory sync, WhatsApp pairing, roles and accounts, and this page. They also include the most basic one, **Talk to the agent at all**, which Viewers do not have by default. ## Data permissions Data permissions are scoped. For example, **See tasks** set to **their own** for a role means people with that role see only tasks they started; set to **everyone**, they see the whole team's. The same scopes apply when Stevin acts on someone's behalf: it reaches only what the person asking is allowed to reach. A Member who may read only their own transcripts cannot get Stevin to read a colleague's. ## What Stevin may decide in a conversation Two settings control whether Stevin asks before it sets something up from a conversation: - **Let a conversation install and connect an MCP** - **Let a conversation switch on a tool of an installed MCP** | Value | Behaviour | |---|---| | ask every time | A confirmation each time. The default for every role | | ask for anything new | Asks only for something that has not been approved before | | never ask | Does it without a confirmation | These settings only remove a confirmation. Who may install or review at all is still decided by the console permissions above. A rejected request is not remembered: the next request asks again. ## Recent changes The bottom of the page lists **Recent changes**: who changed which permission, when, and what the value was before. Use it to review or undo a change. ## Change a permission 1. Open **Permissions**. 2. Find the row and the role's column. 3. Choose the new value. It is saved at once. ## Advice - Change one thing at a time and check that the result is what you meant. - Widen data scopes with care: **everyone** on a read permission lets that role, and Stevin on its behalf, see the whole team's data of that kind. - Keep delete permissions narrow. Deletes always need approval as well; see [Approvals](../features/approvals.md). For every permission and its default per role, see the [Permission reference](../reference/permissions.md). --- path: manual/customize/usage-and-plans.md title: Usage and plans summary: See how many credits the workspace has used this billing period, by day and by model, compare the Free, Pro and Max plans, and manage the subscription. --- ## What credits are Stevin meters usage in credits. Everything it does uses an AI model and consumes credits accordingly: the more steps a task takes and the stronger the model, the more it consumes. Each plan includes a number of credits per month. The allowance resets at the start of each billing period. ## See usage Open **Settings → Plan and usage** in the console. You need the **See usage and billing** permission, held by Admins and the Owner by default. - **This billing period**: the current plan, the period's dates, how many of the monthly credits are used, and when the allowance resets. - **Daily usage this period**: credits used per day, useful for spotting an unusual peak. - **Usage by model**: how many credits each model accounted for, and its share. ## Plans There is one subscription per workspace, and you can cancel at any time. | Plan | For | Monthly allowance | Includes | |---|---|---|---| | Free | Trying Stevin on real work | 1,000 credits | The full assistant, all integrations, background tasks and automations | | Pro | A team getting real work done | 10,000 credits | Everything in Free, plus priority support | | Max | Teams that run on Stevin | 100,000 credits | Everything in Pro, plus priority support | Prices are shown at checkout and on the [website](https://stevinhq.com). ## When credits run out When the allowance for the period is used up, Stevin is paused until the allowance resets or you upgrade. The Usage page says so and shows the reset date. When you are close to the limit, look at **Daily usage this period** first to judge whether it is normal growth or an automation firing too often. ## Upgrade, downgrade and manage the subscription | Action | How | When it takes effect | |---|---|---| | Upgrade | Choose **Upgrade** on the plan and complete checkout | Immediately. The charge is prorated for the days left, and your billing date stays the same | | Downgrade | Choose **Downgrade** on the plan | At the end of the current period | | Payment method, invoices, cancellation | Choose **Manage subscription** | As shown in the billing portal | Starting a checkout or opening the billing portal is its own permission, held by Admins and the Owner by default. If the page says usage pricing needs verification, the totals shown exclude charges that are still pending; contact your workspace Owner. ## Ways to use fewer credits - Set a run limit on [automations](../features/automations.md#run-settings), and write "do nothing when there is nothing new" in the instruction. - Split large, vague tasks into clear small ones to cut wasted steps. - Write the things you do often as [skills](../features/skills.md), so Stevin takes fewer detours. - Keep a lighter [default model](models.md) for everyday chat and use a stronger one only where it matters. --- path: manual/customize/settings.md title: Your settings summary: Personal settings in the console: sign-in credentials, linked chat identity, WhatsApp pairing, phone, voice and microphone, language and appearance. --- ## Where your settings are Open the account menu at the bottom left of the console. It holds **Appearance**, **Language**, **Log out**, and the way into **Settings**. Everything on this page applies to you only; workspace-wide settings are in the rest of [Customize](index.md). ## Account **Settings → Account** shows how you sign in and how Stevin knows it is you. | Item | What you can do | |---|---| | Email | The address you sign in with | | Password | Set or change a password. Changing it signs you out everywhere, and you sign in again with the new one | | Role and plan | Your role in the workspace, and the workspace's plan | | Slack sign-in | Link your Slack account once to use **Sign in with Slack** afterwards | | Chat identity | Whether your console account is linked to you in the chat tool. Without the link, Stevin cannot find you in chat | | WhatsApp | Generate a pairing code to link your own phone. See [WhatsApp](../channels/whatsapp.md#pair-your-phone-each-member) | If you were given a temporary password, the page asks you to set your own before anything else. ## Phone Link the number you will call Stevin from: enter it with its country code, choose **Send verification SMS**, and enter the code within ten minutes. **Unlink phone** removes it. See [Phone](../channels/phone.md). ## Voice **Settings → Voice** controls how [Live](../channels/live.md) uses your microphone and speakers: | Setting | What it does | |---|---| | Microphone | The device Live listens on: the system default or one you pick. Device names appear once the browser is allowed to use the microphone. A change applies from the next Live session | | Hi-fi by default | Starts sessions with the full-band microphone and echo cancellation off. Use only with headphones | | Lower Stevin's voice the moment you speak | Makes it easier to interrupt | | Microphone permission | Shows whether the browser has allowed, blocked or not yet asked for the microphone | If the microphone is blocked, allow it in the browser's site settings for the console. ## Language and appearance In the account menu: - **Language**: the language of the console interface. It does not change the language Stevin replies in; Stevin answers in the language you write or speak. - **Appearance**: **Light**, **Dark** or **Use system setting**. ## Log out Choose **Log out** in the account menu. To sign out of every device at once, change your password. --- path: manual/troubleshooting/index.md title: Troubleshooting summary: What to check when Stevin does not reply, cannot use a tool, keeps asking for approval, an automation does not run, or you cannot sign in. --- ## Stevin does not reply Check in this order: 1. **Did you mention it?** In a channel or group you must mention Stevin. A direct message needs no mention. 2. **Is it in the conversation?** In Slack, invite it to the channel with `/invite @Stevin`. In Feishu or Lark, add it to the group. 3. **Is the chat tool still connected?** Under **Customize → Integrations**, the chat tool should show as connected and online. For Feishu or Lark, the app must also be published and approved; see [Feishu and Lark](../channels/feishu-lark.md#connect). 4. **Is the allowance used up?** When the period's credits run out, Stevin pauses. See [Usage and plans](../customize/usage-and-plans.md#when-credits-run-out). 5. **Does your role allow it?** Viewers cannot talk to Stevin by default. See [People and roles](../customize/people-and-roles.md). ## Stevin says it cannot use a tool or has no permission | What it says | Likely cause | Fix | |---|---|---| | The tool is not connected | The service was never connected | Connect it under [Integrations](../customize/integrations.md) or [MCP servers](../customize/mcp-servers.md) | | The tool is not available | The tool was not allowed in the review, or it changed and is waiting for review | An admin [reviews the tools](../customize/mcp-servers.md#review-the-tools) | | The outside system refused | The connected account lacks access there, or authorization expired | Reconnect and grant the needed scope | | You are not allowed | Your role does not include the action, or the data belongs to someone else | An Owner changes [Permissions](../customize/permissions.md) | ## Stevin keeps asking for approval That is the default: every write is asked about. To be asked less, see [Being asked less often](../features/approvals.md#being-asked-less-often). Deletes are always asked about. ## An approval is stuck A request waits until someone decides. If you see "Another admin must approve this action", it must be decided by an Admin or the Owner other than the person who asked. Find it in **Activity** under **Waiting on you** or **Needs review**. ## A task failed or never finished Open the task in **Activity** and read its result and error. The usual causes and fixes are listed in [Tasks and activity](../features/tasks.md#when-a-task-does-not-finish). ## An automation does not run 1. Open the automation and look at its recent runs. A run that started and failed points to the instruction or a tool; no run at all points to the trigger. 2. **Schedule**: check the time. It is shown in your browser's time zone. 3. **Event**: check that the integration it depends on is still connected, and for webhooks that the sender uses the current key and that the filter matches the body. 4. **Run limit**: runs over the limit are skipped. 5. **Waiting for approval**: a run that reaches a write stops for approval. See [Approvals in automations](../features/approvals.md#approvals-in-automations). Use **Run now** to test the instruction separately from the trigger. ## Cannot sign in | Problem | Fix | |---|---| | The code does not arrive | Check the spam folder and the spelling of the address, wait a minute and request a new one | | The code is rejected | Codes expire after 10 minutes. Request a new one and use the latest | | "Sign in with Slack" says to link first | Sign in with your email code, then link Slack under **Settings → Account** | | Feishu, Lark or Google sign-in fails | An admin must have turned on that sign-in method, and the account's email must match | | The invitation never came | Ask an admin to choose **Resend invitation**, or sign in with the invited address directly | | Your account is disabled | Ask the workspace Owner to enable it | ## Stevin does not recognize me When Stevin treats you as unknown, or the console says your account is not linked, see [Link a chat identity](../customize/people-and-roles.md#link-a-chat-identity). ## Voice and phone See the problem tables in [Live voice](../channels/live.md#problems) and [Phone](../channels/phone.md#problems). ## Contact support If none of this helps, write to [support@stevinhq.com](mailto:support@stevinhq.com). Include: - Your workspace name. - What you asked Stevin to do, and roughly when. - What you expected and what happened instead. - The link to the task in **Activity**, if there is one. Do not send passwords, keys or verification codes. --- path: manual/reference/index.md title: Reference summary: Look-up pages: the glossary, the full permission list with defaults, chat commands, security and data, and how AI agents can read this manual. --- ## Pages in this section | Page | Use it to | |---|---| | [Glossary](glossary.md) | Check what a term means in Stevin | | [Permission reference](permissions.md) | See every permission and its default per role | | [Chat commands](chat-commands.md) | Look up `/new`, `/compact`, `/context`, `/undo` and `/help` | | [Security and data](security-and-data.md) | Understand what Stevin can reach and which controls you have | | [Use this manual from an AI agent](manual-for-agents.md) | Connect an agent to the manual over MCP, or fetch it as plain text | ## Looking for how to do something? Reference pages state facts. For steps, start from [Get started](../get-started/index.md), [Features](../features/index.md) or [Customize](../customize/index.md). --- path: manual/reference/glossary.md title: Glossary summary: Short definitions of the terms used in Stevin and in this manual, from Activity and approval to workspace. --- ## Terms | Term | Meaning | |---|---| | Activity | The console page that lists tasks, their status and results, and the approvals waiting for you. See [Tasks and activity](../features/tasks.md) | | Admin | A role for people who configure the workspace: integrations, MCP servers, team skills, models, usage. See [People and roles](../customize/people-and-roles.md) | | Allowance | The credits a plan includes per billing period | | Approval | A pause in which Stevin describes an action and waits for a person to approve or reject it. See [Approvals](../features/approvals.md) | | Automation | A trigger plus an instruction: Stevin runs the instruction on a schedule or when an event happens. See [Automations](../features/automations.md) | | Background task | Work Stevin does without you waiting in the conversation; you are notified when it is done | | Channel | A place where you talk to Stevin: Slack, Feishu or Lark, WhatsApp, the console, Live, the phone. See [Channels](../channels/index.md) | | Chat identity | The link between your console account and you in the chat tool | | Code execution | Stevin writing and running a short program to compute over data it has. See [Code execution](../features/code-execution.md) | | Console | The web interface of your workspace | | Credits | The unit usage is metered in. See [Usage and plans](../customize/usage-and-plans.md) | | Customize | The console page with the Integrations, Skills and MCP tabs; also the section of this manual on configuring the workspace. See [Customize](../customize/index.md) | | Directory | The list of people in your organization, synced from your chat tool | | Integration | A connection to a chat tool, GitHub or a built-in capability. See [Integrations](../customize/integrations.md) | | Library | The console page with transcripts, summaries and generated files. See [Library](../features/library.md) | | Live | A real-time voice conversation with Stevin in the browser. See [Live voice](../channels/live.md) | | MCP | Model Context Protocol, an open standard for making a service usable by AI | | MCP server | One outside service added through MCP, offering a set of tools. See [MCP servers](../customize/mcp-servers.md) | | Member | The role for people who use Stevin day to day | | Owner | The role responsible for the workspace; the only one that can change roles and permissions by default | | Permission | One thing a role may or may not do. See [Permissions](../customize/permissions.md) | | Plan | Free, Pro or Max; decides the monthly credit allowance | | Role | Owner, Admin, Member or Viewer | | Scope | For data permissions, whose data a role reaches: no one, their own, or everyone | | Skill | Instructions written for Stevin on how to do a kind of job. See [Skills](../features/skills.md) | | Task | One piece of work Stevin does in the background | | Tool | One action an integration or MCP server offers, such as "create an issue" | | Tool review | An admin deciding, tool by tool, which tools of an MCP server Stevin may call | | Trigger | What starts an automation: a schedule, a record change, a GitHub event or a webhook | | Viewer | The role for people who only need to keep informed; cannot talk to Stevin by default | | Webhook | An address another system calls to start an automation | | Workspace | Your team's own Stevin: its people, connections, skills, settings and data | --- path: manual/reference/permissions.md title: Permission reference summary: Every permission on the Permissions page with its default value for Viewer, Member, Admin and Owner. --- ## How to read this page The names are the ones shown on the **Permissions** page in the console. The values are the defaults of a new workspace; an Owner can change every cell except the Owner column. For how the page works, see [Permissions](../customize/permissions.md). ## Console | Permission | Viewer | Member | Admin | Owner | |---|---|---|---|---| | Talk to the agent at all | Deny | Allow | Allow | Allow | | See the team's skills and your own | Deny | Allow | Allow | Allow | | Install Python packages into the code sandbox, and approve installs | Deny | Allow | Allow | Allow | | See connected integrations | Deny | Deny | Allow | Allow | | Connect, disconnect and configure integrations | Deny | Deny | Allow | Allow | | See which tools the agent may use | Deny | Deny | Allow | Allow | | Change which tools the agent may use | Deny | Deny | Allow | Allow | | See the MCP directory and what is installed | Deny | Deny | Allow | Allow | | Install, connect, disable and uninstall a resource | Deny | Deny | Allow | Allow | | Decide which of a server's tools the agent may call | Deny | Deny | Allow | Allow | | Change which packages install without approval | Deny | Deny | Allow | Allow | | See the account behind a person (email, role, status) | Deny | Deny | Allow | Allow | | Re-sync the organization directory | Deny | Deny | Allow | Allow | | Send a member their WhatsApp pairing code, and unlink a wrong number | Deny | Deny | Allow | Allow | | See usage and billing | Deny | Deny | Allow | Allow | | Start a checkout or open the billing portal | Deny | Deny | Allow | Allow | | Switch models | Deny | Deny | Allow | Allow | | Change roles and enable or disable accounts | Deny | Deny | Deny | Allow | | Edit the Permissions page | Deny | Deny | Deny | Allow | ## Data and the agent Values are **no one**, **their own** or **everyone**. | Permission | Viewer | Member | Admin | Owner | |---|---|---|---|---| | See people in the directory | their own | everyone | everyone | everyone | | See tasks | their own | everyone | everyone | everyone | | Create and cancel tasks | no one | their own | everyone | everyone | | Approve what a task is waiting to do | no one | their own | everyone | everyone | | Read a record in full | their own | everyone | everyone | everyone | | Change records | no one | their own | everyone | everyone | | Delete records | no one | their own | everyone | everyone | | Read transcripts | their own | their own | everyone | everyone | | Save sandbox files | no one | their own | everyone | everyone | | Publish transcripts | no one | their own | everyone | everyone | | Create and edit skills (their own: the ones you wrote for yourself; everyone: the team's too, and publishing) | no one | their own | everyone | everyone | ## What Stevin may decide in a conversation Values are **ask every time**, **ask for anything new** or **never ask**. | Setting | Viewer | Member | Admin | Owner | |---|---|---|---|---| | Let a conversation install and connect an MCP | ask every time | ask every time | ask every time | ask every time | | Let a conversation switch on a tool of an installed MCP | ask every time | ask every time | ask every time | ask every time | ## Rules that no setting changes - The Owner's permissions are fixed. - A delete or other irreversible action always pauses for [approval](../features/approvals.md). - An action escalated beyond the person who asked must be approved by an Admin or the Owner other than that person. --- path: manual/reference/chat-commands.md title: Chat commands summary: Commands you can send Stevin in any chat to manage the conversation itself: /new, /compact, /context, /undo and /help. --- ## How commands work A message that is exactly one command is handled as a command instead of being passed to Stevin as a request. Commands work as plain text in every chat channel. In Feishu and Lark you can also type `/` and pick one from the command menu. A command affects only the conversation it is sent in: your own direct-message conversation, or the group thread you sent it in. It never touches anyone else's conversation. Nothing is deleted, and files are untouched. ## The commands | Command | Also accepted | What it does | |---|---|---| | `/new` | `/clear`, `/reset` | Starts a fresh conversation. The old one is archived; `/undo` brings it back | | `/compact` | | Folds the earlier turns into a summary and carries on. You can add what to keep, for example `/compact keep the pricing` | | `/context` | `/status` | Shows what this conversation's context currently holds | | `/undo` | `/resume` | Brings back what the last `/new` archived. Send it again to swap back | | `/help` | | Lists these commands | Only `/compact` takes text after it. "`/new` please" is treated as a normal message. ## When to use them - **Changing topic**: send `/new` so the earlier discussion does not influence the next task. - **A long conversation getting slow or losing detail**: send `/compact`, with a note on what must be kept. - **Started over by mistake**: send `/undo`. See also [Memory](../features/memory.md). --- path: manual/reference/security-and-data.md title: Security and data summary: What Stevin can and cannot reach, the checks before data is changed, what is recorded, and the controls admins have. --- ## What Stevin can reach Stevin does not have unlimited access. What it can reach is decided by three things together: 1. **What is connected.** It can reach only systems that are connected and authorized under [Integrations](../customize/integrations.md) or [MCP servers](../customize/mcp-servers.md). 2. **What is reviewed.** Every tool of an MCP server must be allowed by an admin. Stevin does not use tools that have not been allowed, and a tool that changes comes back for review. 3. **Who is asking.** When acting for a person, it is bound by that person's role and [permissions](../customize/permissions.md). In chat tools, it sees only the channels or groups it has been added to and the direct messages sent to it. ## Each workspace is separate Each team has its own workspace. Its data, integrations, skills and settings belong to that workspace only. ## Checks before data is changed | Kind of action | Rule | |---|---| | Read | Runs directly once the tool is allowed | | Write | Asks a person first by default; admins can adjust this per tool | | Delete and irreversible | Always asks a person; cannot be made automatic | See [Approvals](../features/approvals.md). ## Code execution Programs Stevin writes run in a sandbox with no network access. Installing a package into the sandbox needs approval unless an admin has listed it. See [Code execution](../features/code-execution.md). ## Credentials App credentials and secrets you enter in the console are stored on the server and are not shown again after saving. Webhook keys can be rotated at any time. ## What is recorded - Every task's goal, status and result are in [Activity](../features/tasks.md). - Approval requests and their outcome are part of the task they belong to. - Every permission change is listed under **Recent changes** on the Permissions page, with who made it and the previous value. - Skill revisions are kept. ## Controls available to admins | What you want to control | What to use | |---|---| | Which systems Stevin can reach | Connect and disconnect integrations and MCP servers | | Which actions it can use there | Tool review on each MCP server | | Which actions must be asked about first | **Asks first** or **Runs directly** per tool, and the conversation settings on the Permissions page | | Who can configure all this | Roles and permissions | | Who can see whose tasks, records and transcripts | Data scopes on the Permissions page | | Which tools an automation can use | The automation's MCP server selection | | Who can sign in | Invitations, sign-in methods, and disabling accounts on the People page | ## Calls and recordings Live sessions and phone calls are recorded and transcribed, and kept in the [library](../features/library.md). Phone calls are accepted only from numbers that members have linked and verified. Where the law in your region requires it, tell other participants that a conversation is being recorded. ## Reporting a security issue If you find a security issue, write to [support@stevinhq.com](mailto:support@stevinhq.com). Do not include passwords or keys in the message. --- path: manual/reference/manual-for-agents.md title: Use this manual from an AI agent summary: Connect an AI agent to the Stevin manual over MCP to search it in natural language and read exact pages, or fetch the whole manual as plain text. --- ## What is available | Access | Address | Use | |---|---|---| | MCP server | `https://manual.stevinhq.com/mcp` | Search and read from an agent. No key is needed | | Page index | [`/llms.txt`](https://manual.stevinhq.com/llms.txt) | One line per page: title, link and summary | | Full text | [`/llms-full.txt`](https://manual.stevinhq.com/llms-full.txt) | The whole manual as plain text, with screenshots replaced by their descriptions | | Status | [`/healthz`](https://manual.stevinhq.com/healthz) | Whether the service is up, and what it has loaded | ## Connect over MCP Add the server to your MCP client. Reading needs no key: ```json { "mcpServers": { "stevin-manual": { "url": "https://manual.stevinhq.com/mcp" } } } ``` When an agent connects, it receives a map of the manual (each page's path, title and summary), so it can often go straight to the right page. ## Tools | Tool | What it does | Main arguments | |---|---|---| | `search_manual` | Finds the pages that answer a question | `query`, `limit`, `path_prefix`, `lang` | | `read_page` | Returns the exact text of a page or of one section | `path`, `heading`, `lang`, `raw` | | `list_tree` | Lists the pages under a directory with their summaries | `path`, `depth`, `lang` | ## Search in natural language `search_manual` takes a question, a symptom or a few keywords, in any language. The wording does not have to match the manual: - "how do I stop it asking for approval every time" - "bot not replying in Slack" - "who can change permissions" It returns up to `limit` pages (8 by default), each with its best-matching section: `path`, `heading`, `anchor`, `url` and a `snippet`. Use `path_prefix` (for example `customize`) to search one section of the manual. Search finds the page. For exact wording, follow up with `read_page`. ## Read exact content `read_page` returns the page's metadata (`title`, `summary`, `tags`, `aliases`), the list of its headings, and its Markdown `content`. - Pass `heading` to get one section only. - Pass `raw: true` to also get the complete source file, including the front matter. - Paths are forgiving: `customize/permissions`, `manual/customize/permissions.md` and an old path listed in a page's `aliases` all resolve to the same page. A directory name returns its index page and the list of its children. ## Screenshots are readable too Every screenshot in the manual is written as a figure with numbered callouts. `read_page` returns them under `figures`: the image address, a description of the image, and for each callout its position, label and explanation. An agent that cannot see images gets the same information a person reads in the legend under the screenshot. ## Languages The manual is written in English. `search_manual` understands questions in other languages, and the `lang` argument selects a translation where one exists. When a page has not been translated, the English page is returned and the result says so. ## For editors of the manual People who maintain the manual connect with an admin key and get additional tools for drafting changes. Drafts are reviewed by a person before they are published. This is described in the manual's repository, not here.