Every AI coding tool on your team, one set of rules.
Your team’s coding rules and decisions, approved by people and kept in one shared vault. Cursor, Claude Code, Codex and every other MCP client read it automatically, so each developer’s AI works from what the team actually agreed.
$39 per seat, per month. Self-serve up to 20 seats.
Every developer’s AI follows different instructions.
Rules drift
Standards live in copied AGENTS.md and .cursorrules files, one per repo, each a little different and out of date within weeks.
Decisions get relitigated
What the team settled lives in people’s heads and old threads, so the AI keeps suggesting the approach you already rejected.
No one can say what it was told
When someone asks which instructions the AI followed, and who approved them, there is no record to point to.
Two vaults, kept apart.
Joining a team moves nothing you have written privately. The two are read separately, so a team feed shows the team’s memory and nothing else.
Your personal vault
Everything you capture for yourself, in any tool you connect.
- Visible only to you
- Unchanged when you join or leave a team
- Never surfaced to teammates or their agents
The team vault
What the team has agreed, written deliberately into the team.
- Readable and writable by every member
- Each write carries who made it
- Reaches agents only once approved
Three steps, no admin setup.
Create the team
Name it, and you are its owner. The team gets its own feed, separate from your personal one, and a switcher to move between them.
Invite with one link
Share the invite link, or the code it carries, like CX-A3F7-K2Q9-4M. Anyone who has it joins and can read every memory shared inside, so pass it on the way you would a key to the room. The owner can regenerate it at any time.
Write to it
Add a memory from the dashboard, from your editor over MCP, or by importing what another tool already knows. It lands in the team, attributed to whoever wrote it.
A write is not yet a rule.
Anything an agent writes to the team, and anything extracted from a chat, arrives pending. A member reviews it and approves or rejects it. Until then it sits in the queue and governs nothing. What a member writes themselves goes straight in.
Only approved memories are returned to your agents and put into their context. That is what the queue is for: an unreviewed write cannot quietly become a rule the whole team’s agents follow. Every member can review, so the queue never waits on one person, and only a person can: an agent’s key can propose, never approve.
Two roles. No admin tier.
Deliberately small, so the answer to “who can do this” is always one of two names.
| Role | Can |
|---|---|
| Member | Read the team vault, write to it, review what is pending, and leave the team. |
| Owner | Everything a member can, and remove a member, regenerate the invite code, or delete the team. |
The same vault, over MCP.
Connect your editor once and the team vault is there. Three tools do the team work, and the team is named in every call, so nothing is shared by accident.
list_my_teamsThe teams you belong to and your role in each. Where the team id comes from.
get_team_memoriesThe approved shared memories, filtered by type if you want. Pending ones are not returned.
add_team_memoryWrite to the shared vault. Membership is checked against your own credential, never a name supplied in the call.
Every write is attributed. The person and the tool come from the credential that made the call, and cannot be forged. An agent that names itself is recorded as having said so, and marked unverified, because a name a caller chooses for itself is a claim rather than a fact.
An 8-week pilot, shaped around your team.
We’re working with a small number of engineering teams to build the parts a larger org needs. You get the team vault working across your developers’ tools now, and a say in what comes next.
Ready on day one
- Shared team vault and review queue
- MCP set up for every developer’s editor
- Your existing rules and decisions brought in
- A weekly check-in with us
Built with you during the pilot
- An org admin console across teams
- A log of what context each tool was given, and when
- Single sign-on when you need it
Give your team’s AI one set of rules.
$39 per seat, per month. Unlimited memories, the shared vault, the review queue and roles. Self-serve up to 20 seats; for a larger team, run a pilot with us.