Your company’s AI brain
for people & agents

An AI brain is a single, governed place where a company's documents, conversations and tools become shared knowledge that both people and AI agents can ask questions of. Brain is that layer for your company. It connects your sources, keeps the permissions each source already enforces, answers questions with the source cited, and records every access in an audit log.

There is one shared knowledge layer for the whole company, and each person's view of it is shaped by the permissions they already have. Sales and finance ask the same Brain the same way; they get answers built only from what each of them could already open. Nothing has to be copied into a separate workspace per team, and no one is handed access they did not already have.

Yes, through MCP. Claude, Cursor and Codex connect to Brain today, along with other clients that speak MCP. An agent gets persistent context without reloading a conversation history into every session, and it is governed like a person: its own identity, its own scoped key and a mandate that limits what it can reach.

Four controls, all of them live today. Identity: every request is tied to a named person or a named agent. Scope: an agent's key limits what it can reach, independently of what its owner can reach. Policy: answers are assembled only from permitted sources, and what was withheld is noted. Audit: every access is written to a content-blind record of who asked, what was reached, what was withheld and when. An agent can be revoked, and the revocation applies on its next request.

Single sign-on, permissions inherited from each source rather than re-declared in Brain, a content-blind audit record of every access, and admin control over which agents exist and what each one can reach. Brain is hosted by us: your data is yours, but the deployment is not something you run yourself. Plan limits differ, so check the plan you are on for the number of sources, seats and agents it covers.

They solve different problems. Obsidian is a local-first notes app: your notes are files on your machine, it has a large plugin ecosystem, and teams share through synced vaults and shared files. Brain is governed shared context for people and AI agents: it reads the sources your company already uses, answers questions from them with citations, applies each source's existing permissions per person, and logs every access. If what you want is a team notes app, Obsidian is the better fit. If what you want is one place your team and your agents can ask questions of, that is Brain.

Live today: Google Drive and Notion as sources, with their permissions intact, and Claude, Cursor and Codex as AI clients over MCP. On the roadmap, not yet available: Slack, Gmail, Box, Telegram, GitHub, Confluence and Salesforce. Everything that ships goes through the same governed path, so a new source does not mean a new set of access rules.

No. Brain uses governed retrieval only, and nothing you connect is used to train any model of ours. The model that writes the answer is a third party, so its own terms apply to the text of a request; you can bring your own model key, in which case the request runs under your agreement with that provider rather than ours.

Sign up with your work email, connect one source, and Brain starts building context from it. Permissions come across with the content, so nobody sees anything they could not already open. There is a free plan with a monthly allowance of credits, which are spent on queries and on indexing new content; paid plans raise that allowance and add more sources, seats and agents.

Permissions are inherited from the source, not declared again in Brain, so a document you cannot open in Drive is a document Brain will not answer you from. Agents are scoped separately from the person who created them and can be revoked, which takes effect on the agent's next request. Every access is written to a content-blind record: who asked, what was reached, what was withheld and when. The record holds the access, never the content, and it is independently checkable by your own auditor.