Sedimenta
Where knowledge settles.
Every value is threaded back to its source. Agents propose, people approve — and what’s verified settles into your team’s memory, on your own server.
| Record | Field | Value | Source | Status |
|---|---|---|---|---|
| Company A | revenue 2025 | $48.2M | annual_report.pdf | verified |
| Company A | headcount | 120 | company website | verified |
| Company B | round | Series A | press_release.html | proposed |
| Company B | revenue 2025 | $31.5M | annual_report.pdf | verified |
| Deal D | amount | $180,000 | client_email.eml | proposed |
| Deal D | start | November | client_email.eml | verified |
Agents propose. People decide. Knowledge settles.
An agent connects over MCP and reads the rules of the space. A write that breaks a required rule is refused; where you ask for review, the agent writes a proposal with its source. Once a person approves it, the value settles with its thread — and every agent that reads it later gets the provenance too.
Everything your agents know, your people can see. And it follows your rules.
A source on every field
Pages, databases and records that show where each value came from, who proposed it and who approved it.
Inbox and contradictions
Agent proposals, conflicting sources and expired claims in one queue — resolved in one step.
History and “as of” any date
Every before-and-after change, and what the system knew on any day in the past.
What’s not allowed
Required rules are enforced by the server. A violation doesn’t get through — from a person or an agent.
What’s better
Guidance matched to the task that warns instead of blocking. Try any new rule in report-only mode first.
What agents can do here
Ready-made commands attached to a space, database or record — served over MCP, and as a plugin in Claude Code.
Everything you expect from a modern workspace
Spaces, nested pages, a block editor and databases with views — the workspace your team expects, with people and agents working side by side. Not a black box behind an API.
An open core. Extend it with modules.
Add your own record types, screens, calculations and MCP tools as a module on the SDK — your coding agent can write one, too. Meanwhile agents extend the workspace as data: types, formulas, rules and skills, no deploy needed.
Modules on the SDK
Agents extend as data
Module catalog coming soon
Sandboxed plugins coming soon
From one command to your team’s memory
Connect an agent
Run it on your laptop and connect Claude Code, Cursor or Codex.
npx sedimentaInvite the team, set the rules
A server in one command, spaces, permissions and your first rules for agents.
docker compose up -dA verified knowledge base
Agents turn documents into records with sources, people approve, and knowledge keeps settling.
create_proposal → acceptYour server. Your data. Your model.
Your server
One Docker command, or a desktop install for one person.
Your data
PostgreSQL, files on disk or S3, backups you control.
Your model
Any model provider, any MCP client, your own keys.
No lock-in
Export everything to readable markdown at any time.
Free, forever
Self-host with no limits on people or agents. Provenance, rules, MCP, permissions and full export — free, for good.
No per-seat pricing
Cloud priced per workspace, SSO and audit for organizations, document extraction priced by usage.
An open benchmark for memory integrity
How many wrong values get into the base when agents write — with the method and code to reproduce it.
Get notifiedQuestions
How is this different from Notion?
It runs on your own server, every value carries its source and history, and AI agents write through proposals that people approve — under rules the server enforces.
How is it different from Mem0 or Cognee?
Those are memory APIs and engines for developers. Sedimenta is a workspace people live in, with provenance, permissions and approval in the open core.
Where is my data stored?
In PostgreSQL on your server, or a local database on your laptop. Everything exports to readable markdown files.
Which models and agents work with it?
Any MCP client that connects with a token — Claude Code, Cursor, Codex, VS Code, Gemini CLI. One-click remote connectors (ChatGPT and others) are coming. Sedimenta is neutral to the model provider.
What does it cost?
The self-hosted edition will be free with no limits on people or agents. Cloud and organization features are coming — request early access to hear first.
Is it open source?
Sedimenta is in private preview. We’ll publish the repository and announce the license before the public release.
Request early access
We’re working with a small number of teams that already use AI agents in their daily work.
hello@sedimenta.ai