How it works
Your assistant works in Tavimi. You follow along from your browser.
Tavimi is an MCP server. You connect your assistant once; from then on, it reads and writes your project notes and tasks in every conversation. Your team sees the same work in the web app, and answers the questions it asks.
The Tavimi web app is in French for now; an English interface is on its way. Your assistants work in any language.
Two ways in, one shared memory
Your AI assistants
- Claude: claude.ai, Desktop, Code
- ChatGPT and Codex
- Mistral's Le Chat
- Any MCP client
Each one signs in with your account, to the spaces you choose.
MCPread and write
One space per project
- Notes
- Tasks
- Documentation
- Files
- Questions
Nothing is overwritten, nothing is deleted: every edit keeps the previous version.
Web and phonefollows and answers
Your team
- The task board
- Activity, day by day
- Questions to settle
- Search across everything
Everyone writes too: notes, tasks, answers.
Get started
Three steps, and your assistant has a memory
Create your account
An email address, a password, then a second factor: an authenticator app or a passkey. It's required for everyone.
Create one space per project
A client, a campaign, prospecting: a name is all it takes. Invite the people you work with by email; they only see that space.
Connect your assistant
Add Tavimi as a connector. On the first call, you sign in and tick the spaces it will be able to see.
claude.ai, Claude Desktop, ChatGPT and Le Chat: connector URL
https://app.tavimi.com/mcpClaude Code
claude mcp add --transport http tavimi https://app.tavimi.com/mcp
What the assistant files
Notes, tasks, documentation, files: the project's memory
Notes that don't fade
A decision and the reason behind it, a pitfall you hit, a useful resource. Each category has its own writing rules: the assistant follows them, and an incomplete note is rejected.
- decision
- pitfall
- reference
- your own
Tasks you can track
Status, priority, due date, and an owner: a person, or a role such as “copywriting”. The assistant first picks what isn't waiting on anyone's answer.
- Write the launch postscopywriting
- Approve the main visualoverdue
Documentation that stays current
The assistant consolidates notes and completed tasks into a summary document: project status, market watch, communication plan, sales pipeline. It knows what it has already read, and only picks up what has changed.
Files stored with everything else
A report, an export, an image, up to 10 MB: the assistant or the team uploads them to the space. Uploading a file with the same name creates a new version.
Search across all your spaces
For you and for the assistant, regardless of accents or word forms: “decided” finds “decision”.
A project timeline
If you turn it on, the assistant keeps the space's timeline, week by week: what was shipped, decided and learned, with a link to every note or task it cites.
Questions
An assistant that asks instead of guessing
When it's missing a choice, the assistant asks the question on the task, with its options and the one it recommends, then moves on to another task. It doesn't get stuck, and it doesn't make things up.
- You answer when you have time, from your browser or your phone, where Tavimi installs like an app. All the questions on a task are sent in one go.
- For each space, you say which decisions must always come back to you: the assistant receives that rule with every task it reads.
- The decision stays yours: you answer in Tavimi, or you dictate your answer to your assistant, which only saves it once you confirm.
- Every answer is logged on the task, under your name, and can still be edited. The assistant reads it and picks up again.
Write the launch posts? 1
How many posts per week during the launch?
Add details or answer differently
A typical session
You say “pick up the campaign.” Here's what happens.
It reads the space
The project description, the decisions made, the open tasks and the answers to its questions.
It picks a task
The highest-priority one that isn't waiting on anyone, and it moves it to “in progress”.
It works and files
It logs progress on the task, writes its decisions as notes, and asks a question if a choice is up to you.
You see the result
The completed task shows up in the space's Activity, with a badge until your next visit.
You stay in control
The assistant only sees what you open to it
Spaces you choose. When it connects, you tick the spaces, or whole organizations, the assistant can see. The rest stay out of its reach.
Access cut in one click. The “Mes connexions” (My connections) page lists your assistants' connections. Revoking one cuts it off without touching the others.
Two-factor authentication required for everyone, in the web app and when connecting an assistant.
Full history. Every version of a note or task can be viewed, with its author. Nothing is deleted.
FAQ
What people ask us
What is an MCP server?
MCP (Model Context Protocol) is the standard way to connect a tool to an AI assistant. Once Tavimi is added as a connector, Claude reads and writes in it without any copy-pasting, over a connection secured by OAuth.
Do I need to connect each assistant separately?
Yes: each connection is separate, with its own spaces. But a single connection covers all the organizations in your account.
What happens if two assistants edit the same note?
The second one is told the note has changed: it reads it again and reapplies its edit. Nothing gets overwritten.
Can I connect Tavimi to n8n, Make or a script?
Yes, through its REST API. You create a personal API token under “Mon compte” (My account): it acts on your behalf, with your permissions, is limited to the spaces or organizations you choose, read-only if needed, and can be revoked at any time. It reads and writes tasks and notes, reads documentation, and uploads files.
Can Tavimi see what happens in my terminal?
No. Tavimi only receives what the assistant writes to it.
Ready to connect your assistant?
Create your account and your first space, then connect your assistant: it won't start from scratch again.