Workspace and people
| Term | Meaning |
|---|---|
| Workspace | Holds your projects, members, AI provider Credentials and billing. Every project belongs to one workspace. Members and roles |
| Overview | A workspace's page of what its tasks got done, where they ran and what waits on you. Open it from the workspace's … menu in the sidebar. Overview, Needs you and Activity |
| Files | A shared drive of folders and files for a workspace, and a personal one for you. It also holds what agents write, in Sessions and Crons folders, and a session's or Cron run's Files tab lists its own. Agents are told not to delete. Files and Files of a session |
| Member | A person who belongs to a workspace. Every member can open and send into every chat session in it. Members and roles |
| Role | What a member may do in a workspace. It is Owner, Admin or Member, and owners and admins manage members, Credentials and settings. Members and roles |
| Credentials | The workspace's stored access to an AI provider, such as an API key or a Claude Code or Codex login. Admins manage them. AI providers |
| Subscription | Your own Claude Code, Codex, Antigravity, Muse Code, GLM Coding Plan, Kimi Code or DeepSeek API access. It runs personal work while you hold any; Sensei free is offered only when you hold none. Workspace work uses workspace Credentials only. AI providers |
| Provider | A company whose models Sensei runs, shown by its product: Claude, Codex, Gemini, Muse, GLM, Kimi or DeepSeek. AI providers |
| GitHub Connection | Your link to your own GitHub account. It lists the repositories you can import and puts your name on the commits and pull requests from your messages. Import a repository |
| Kosuke App | The GitHub App you install on a repository so Sensei can read and push its code. It opens pull requests for members without a GitHub Connection. Import a repository |
| API Key | Your own key for the Sensei CLI, the MCP server and the API. It acts as you. Authentication |
| Connector Token | What an AI app such as Claude or ChatGPT holds once you allow it on Sensei's consent screen. It acts as you, like an API Key, and stops working when you disconnect the app. Connect your coding agent |
| Connected App | An AI app you have allowed to act as you through the Sensei MCP server. Settings lists it next to your API Keys, and you disconnect it there. Connect your coding agent |
| Sensei free | Runs personal work when you hold no Subscription, like a personal Subscription within the Free allowance and staff maximum Turn Level. It covers Sensei, personal tasks, personal-project Deploys, Personal Crons with or without repositories, and Delegated Runs. Its model, Source and provider stay hidden. AI providers |
| Free Credits | Historical USD workspace grants, retained separately from Personal Free’s one-time actual-hour allowance. Plans and billing |
Projects and deploys
| Term | Meaning |
|---|---|
| Project | A GitHub repository imported into a workspace, or made from kosuke-starter, plus everything Sensei needs to run it. Import a repository |
| Deploy | One attempt to get the project's app running in its main sandbox. A redeploy tries the saved settings first, and the deploy agent works out or fixes how to run the app when needed. Importing a repository starts the first one, and a project needs one successful deploy before sessions can start. Deploy your project |
| Deploy Chat | The project's conversation with the deploy agent, in the project's Settings view. Deploy your project |
| Sandbox | The cloud machine a project or a session runs in. It costs money while awake and pauses when nobody uses it. Preview and sandboxes |
| Preview | Your app running in a chat session's sandbox, shown inside Sensei and at its own web address. A repository with the kosuke-starter layout also has a Preview page, which shows its default branch. Preview and sandboxes |
| Target Branch | The branch a session's pull request targets and a Sync brings in. Unless a project sets one, it is the repository's default branch. Project settings |
| Version | One merged task, or one Chat push, in a repository with the kosuke-starter layout, which can be published from its Preview. Preview and publish |
| kosuke-starter layout | What Publish needs to build the app: bun, @cloudflare/vite-plugin, a Wrangler config and a Turso database. A repository with it opens on its Preview. Preview and publish |
Chat sessions
| Term | Meaning |
|---|---|
| Chat | The project's one chat agent on its default branch, the third tab of the repository's chat column. It pushes there itself, with no pull request, and is off where GitHub's rules or your access forbid that. Project settings |
| Chat session | One unit of work on a project. Once started, it has its own branch, sandbox and conversation, and later a pull request. Chat sessions and tasks |
| Tasks page | A workspace's page of tasks, one column each for Todo, In progress, Done and Archived. A repository's own page shows its settings, not a board. Chat sessions and tasks |
| Backlog | A session holding only its prompt, ready to start. It has no branch or sandbox yet. Chat sessions and tasks |
| Active | A session whose conversation is open. Its pull request, if it has one, is a draft. Chat sessions and tasks |
| Under Review | A session whose pull request is ready for review. A member's message resumes it, except a sync, and so does Resume. Chat sessions and tasks |
| Completed | A session whose pull request merged, or a Backlog session completed without code work. Chat sessions and tasks |
| Archived | A session set aside without a merge. Its pull request is closed and its sandbox removed. Chat sessions and tasks |
| Assignee | The member a task is booked to. Workspace tasks use workspace Credentials regardless of Assignee. Chat sessions and tasks |
| Turn | One exchange with the agent. It covers your message, the agent's work and its reply. Messages, Up next and steering |
| Up next | The messages waiting behind the running turn. Each one starts its own turn, in order. Messages, Up next and steering |
| Steer | Sending your waiting message into your own running turn, so the agent changes course without starting a new turn. Messages, Up next and steering |
| Clarification | A question the agent asks before it goes on. The turn ends, and your answer starts the next one. Messages, Up next and steering |
| Resume | Turns the pull request back into a draft and moves the session back to Active. Open a pull request |
| Review | /kosuke:review: one model from another family tries to break the change and reports its MAJOR and CRITICAL problems, which the agent fixes. /kosuke:ship runs it before QA. /kosuke:review |
| Sync | Brings the session's branch up to date with the Target Branch. GitHub merges a clean sync; the agent handles conflicts and local work. Sync with the Target Branch |
| Merge agent | One per repository: an agent of Sensei's that merges the pull requests with a go-ahead whose QA passed, as the Kosuke GitHub App, woken by events. Its mode is the merge train (one at a time) or Off (Sensei merges nothing; members merge themselves). An imported repository starts Off; one made from kosuke-starter starts on the merge train with Go-ahead When checks pass. Project settings |
| Needs you | Sessions blocked on a person, for example by failing checks or a question from the agent. Each one reaches its Sensei thread as an event, and Sensei draws a card when it needs your decision. Overview, Needs you and Activity |
| Activity | The workspace's log of what changed and who changed it, kept for 30 days. Overview, Needs you and Activity |
Skills and automation
| Term | Meaning |
|---|---|
| Skill | Instructions the agent follows when you run them, with /name in a Claude, GLM, Kimi or DeepSeek session or $name in a Codex session. Skills |
| Workspace Skill | A skill kept in the workspace's settings. While it is on, every chat session in the workspace has it. Skills |
| Factory Skill | A skill Sensei ships in its kosuke plugin, such as /kosuke:open-pr. Every session carries it. Skills |
| Personal Skill | A skill you keep for yourself. It is loaded into every session you are the Assignee of, in every workspace. Skills |
| Project Skill | A skill stored in the repository itself. Skills |
| Plugin | A bundle the agent installs from a marketplace. It can add skills, commands, subagents or hooks. Sensei offers only plugins that carry no MCP server. Plugins |
| Cron | A saved prompt handed on a schedule to the Cron Agent in the projects it covers, which may be none, all in one workspace. Each fire makes one Cron Run. It runs as the workspace, on the workspace's Credentials. Crons |
| Personal Cron | A cron that belongs to you and to no workspace. With no repository it runs in its own Sensei thread. With your Personal Repositories it runs the Cron Agent in a sandbox of its own. Both use your Subscriptions, or Sensei free while you hold none. Crons |
| Sensei | Your own agent across every workspace you belong to, marked Beta. It has no workspace of its own and runs on your Subscription or Sensei free. You have one. It reads every project and session there and has no modes. It acts on sessions only when you ask, and it draws the cards that need you when the app writes events in a thread. Sensei |
| Prototype | A page Sensei draws to show a change before it is built, saved as an HTML file in the workspace's Files. Variants are sibling files in a folder. Sensei |
| Thread | One topic in your Sensei chat. It holds your replies, the sessions it is about, the events written in it, and what waits on you there. The Threads tab lists them, offers All, Needs you, Working and Done in its filter and counts Needs you. Its dot means Needs you; Mark done sets aside a Clarification or your own working sessions until new activity. Sensei |
| Event | One line the app writes in a thread when something happens, such as "Checks failed on #2161: typecheck, lint". It records and asks nothing, and sends no notification. Sensei |
| Event Turn | A Sensei turn the app starts in a thread about 45 seconds after the last of a burst of events. Sensei reads the events and opens, updates or closes a card, or does nothing. It shows only the card. Sensei |
| Card | One thing a thread asks you to decide, designed by Sensei: a headline, parts such as checks, a QA report or a diff, an instruction, and a negative button (Reject by default, or Sensei's own words), Reply and a button named for what it does. Sensei |
| Receipt | A one-line note the app posts in a thread when something happened that asks nothing of you, or when a command Sensei ran for a card finished. Sensei |
| Message | Something a cron or an agent sends your Sensei for a thing it may not do itself, such as a ticket to file, tickets to merge or a QA guide to save. Sensei draws a card for it, and approving runs it as you. Sensei |
/kosuke:ship | A Factory Skill that carries a ticket inside its session from grounding to a pull request marked ready. It stops only when it needs a person. Sensei |
| Ready | A session where /kosuke:ship stopped with ready. Its checks are green, QA passed or was skipped, and the review passed or was skipped. Sensei |
| Go-ahead | Your go-ahead to merge one pull request at one commit, kept by the app on the session. You give it, or Sensei gives it for you; it counts only at that commit, so a push leaves it on an older head. With the merge train, the merge agent merges it; with the merge agent Off, you merge it yourself. Under the merge train's When checks pass, required checks that pass on a ready pull request's latest commit give it. A GitHub label or auto-merge is no go-ahead. Sensei |
| Merge rule | Your standing rule, set once in words to Sensei, that Sensei gives the go-ahead to each of your pull requests once it is approved and its QA passed. Sensei |
Models
| Term | Meaning |
|---|---|
| Model | The AI a session runs. It combines a provider, a model family, a reasoning effort and, on some Codex models, a speed. You pick it for each session. Models |
| Reasoning effort | How hard the model thinks, such as Low, Medium or High. You pick it as part of the model. Models |
| Speed | How fast a Codex model answers. Standard is the default. Fast and Ultrafast cost more. Models |
| Fast | A paid speed tier, about 1.5x to 2x faster depending on the model. Models |
| Ultrafast | The fastest paid speed tier, for GPT 6 Astra only. It is not the Ultra reasoning effort. Models |
| Default Model | The model a new task starts on when nobody picks one. Models |