What is Sedimenta

Sedimenta is the system of record for what your agents know. It is a self-hosted workspace where people and AI agents share one memory: pages and databases where every value keeps its source, agents write under rules the server enforces, and people approve what becomes true for the team.

The problem it solves

AI agents now draft documents, fill tables and answer questions all day. Without a shared memory, their work is lost in chats, numbers appear without a source, and an agent with write access can overwrite a page with something it made up.

How it works

1. Agents connect over MCP. Claude Code, Codex, Cursor, OpenCode or your own agent connect with a token. Claude.ai and ChatGPT connect through OAuth.
2. They read the rules first. Each space has rules. An agent reads them before writing and passes their fingerprint with every write.
3. The server enforces them. A write that breaks a required rule is refused with the reason. Where you ask for review, the agent leaves a proposal with its source instead.
4. People decide. A person approves or rejects proposals. Approved values settle with their thread: the document, the quote, who proposed it and who approved it.
5. Everyone reads the same memory. People work in the web app, agents over MCP. Any agent that reads a value can ask where it came from.

What is inside

• Pages and databases. A block editor, typed databases with relations, formulas and rollups, table, board, list, gallery and graph views, real-time editing.
• Memory with provenance. A source on every field, a queue of proposals and contradictions, history and the state on any date.
• Rules, recommendations and skills. Required rules checked by the server, guidance that warns instead of blocking, and ready-made commands for agents.
• Permissions. Workspaces, spaces, folders, per-record access and explicit deny. Every agent works with its own token.
• A verifiable journal. Every change with its value before and after, chained by hashes.
• Files as records. Immutable versions on disk or S3, with permissions checked before every download.

Where it runs

On your own server with PostgreSQL and Docker Compose, or on one person’s laptop. Your data stays with you, and Sedimenta can run with no outgoing requests at all. Everything exports to Markdown and the open OKF format.

Status

Sedimenta is in private preview with a few teams that already work with AI agents. Request early access.