Getting started with onecell and your agent
Connect your agent, learn where work goes (Inbox, cells, the hive), run your first recipe, then publish or share the result. About ten minutes.
onecell is memory for your work and your team that your agents can read and write. You keep notes, decisions and results in cells; your agent searches them, cites what it finds, and leaves its own work there for the next person or the next session. This guide gets you from nothing to a first useful result in about ten minutes.
1. Connect your agent
onecell works through MCP, so any agent that speaks it can use it: Claude Code, Claude and Cowork, Cursor, Codex, Grok, ChatGPT and others. Setup gives you the exact command or connector for each. You sign in once in the browser; nothing to copy by hand.
Prefer a key? Create one on Settings → Agents. Choose Drafts only for an agent you want to read and draft, but never publish, share, delete or invite.
2. Know where things go
| Place | Who sees it | Use it for |
|---|---|---|
| Inbox | Only you, in every cluster | Drafts, sessions you park, anything not ready for the team |
| A cell | You, and whoever you share it with | One project, client or topic |
| The hive | Everyone in your cluster | What the whole team should know |
| A cluster | Its members | Your team: its members, its hive and its cells |
When you are not sure, use your Inbox. Moving a document to a cell later is one step, and it brings anything nested under it along.
3. Run your first recipe
Pick one for your job
The recipes are grouped by who you are and what you are doing: developers, marketing, support, research, founders, and automations.
Copy the prompt
Each recipe is one prompt in a box with a Copy button.
Fill in the blanks
Replace the parts in <angle brackets>. Leave out <cell> and the result goes to your Inbox.
Paste it into your agent
It reads what onecell already knows, does the work, and saves a draft.
4. Check the draft, then publish or share it
Everything an agent writes starts as a draft, and every change it makes is tried as a dry run first, so nothing lands by surprise. Read the draft in onecell. When it is right:
- Publish it, so search and other agents treat it as settled. Publishing checks a little more strictly than drafting: a title, a summary and enough text to stand on its own.
- Share the cell with a teammate as viewer or editor, or open it to everyone in your cluster.
- Make a link for someone outside the team. A link is read-only, and you can revoke it.
Every save is a version, so you can always see what changed and who changed it.
5. Bring your team
Invite people from Members. They join your cluster and get the hive, plus any cell you have opened to the whole cluster. Your other cells stay yours until you share them, and your Inbox is never shared.
Good habits
- Ask for the artifact you want by name: a diagram, a chart, a timeline, a decision, a checklist. Your agent builds it as a real block you can edit later. See what it can make in Artifacts.
- Let your tools write here too. GitHub Actions, n8n and scheduled agents can keep their work in onecell: see Automations.
- Ask your agent to cite what it found as cell · heading · passage. A cite you can open is worth more than a confident summary.
- Write decisions as decision blocks. Your agent can list every open one, across clusters, with a link.
- If your agent says a onecell tool is missing, reconnect it (in Claude Code,
/mcp→ onecell → reconnect). onecell shipped something after it connected.