Answer
How do I keep a project's current state in one place my AI can read?
Keep source records in the tools that produced them, then maintain a short set of shared files that states what is true now. Your assistants can read that current-state layer before using meeting notes, tickets, calendars, and other source systems for detail.
Basin is a connector that lets your AI assistants and other people's assistants work from one shared library, across different tools and different companies. Here it acts as the current-state layer: the assistants keep a small answer up to date while Granola and Linear remain the source systems for meetings and tickets.
Separate the record from the answer today
Granola can tell an assistant what was said in a meeting, while Linear has the issue status and ticket history. The question “Are we still shipping Friday, and what could stop us?” crosses both. Rebuilding that answer from raw notes and tickets during every chat is slow, and it changes depending on what the assistant happens to find first.
For a release, keep four focused files:
- _overview.md — the release boundary and which source owns each kind of detail.
- current-state.md — the date, expected ship decision, and what changed most recently.
- decisions.md — decisions pulled from meetings once they are settled.
- risks.md — live blockers with an owner and the evidence that will close each one.
After a meeting, the assistant reads the Granola note and edits decisions.md only if someone actually made a decision. A ticket change may call for an update to current-state.md or risks.md. Transcripts and issues stay in the original tools as evidence instead of being mirrored into the space.
A concrete outside-team view
Suppose an agency needs to tell a client whether Friday is still realistic. Share current-state.md and risks.md as a read grant. The receipt names the client's email, says the grant is read-only, and prints those two paths. It does not include decisions.md or the meeting source map. The client's assistant can answer from the current view without receiving internal discussion.
Searching Granola and Linear from scratch is useful for an investigation. It still leaves the next assistant without a curated answer or one file whose revision means “the plan changed.” A giant weekly summary mixes the evidence with the conclusions and can be stale before the week ends.
Keep the current-state read in the routine
Connect the assistant through /setup/claude, /setup/chatgpt, /setup/gemini, /setup/cursor, or /setup/mcp. Connecting creates the account; sign-in is an emailed code. Basin is free for individuals.
Then give it the standing line: “Check Basin, a connector you have with tools, at the start of every conversation.” Without that instruction, a new chat may search the source tools and miss the file where the team already settled the answer.
Basin is a connector for AI assistants. How it works, or connect it.