sedimentaRequest early access
For agents

Give your agents a memory your team can check.

Sedimenta is an MCP server and a workspace in one. Claude Code, Codex, Cursor, OpenCode or your own agent read and write the same pages and databases your people use — with rules, approval and a source for every value.

Connect

One line in your agent’s config

Any MCP client works: give it the address of your installation and a token. Claude.ai and ChatGPT connect through OAuth instead of a token.

# Claude Code
claude mcp add --transport http sedimenta https://memory.your-company.com/api/mcp \
  --header "Authorization: Bearer $SEDIMENTA_TOKEN"

# any other MCP client: the same URL and header
The loop

Read the rules, write with a source, let a person decide

› get_rules(space="research")
  ✓ 3 rules · fingerprint r1-7f2c
› update_record(type="decision", id="decision-12", patch={status:"done"}, rules_seen="r1-7f2c")
  ✗ 422 rule_violation · a decision must name who decided
› create_proposal(entity_type="company", entity_id="company-1", field="revenue_2025",
      after=48200000, fragment_ids=["fragment-7"])
  ↳ proposal-31 waiting for a person
  ✓ approved by Anna · 14:02
› get_provenance(type="company", id="company-1", field="revenue_2025")
  $48.2M ← annual_report.pdf · p. 4 · fragment-7 · approved by Anna
Rules with a fingerprint

The agent passes the fingerprint of the rules it read. If the rules changed since, the server answers 428 with the new rules, so an agent never writes against rules it has not seen.

Refused, not ignored

A write that breaks a required rule gets 422 with the reason. Recommendations warn without blocking.

Proposals

Where a space asks for review, the agent proposes a value with the fragment of the document it came from. A person approves or rejects it.

Provenance on read

Every agent that reads a settled value can ask where it came from: document, page, quote, who proposed it, who approved it.

Tools

What agents can do over MCP

read

Find and read

Search pages, records and documents with quotes; query databases; read a record with its rules; get a context pack for a task.

search · search_knowledge · query · get_record · context_pack
write

Write

Create and update records in bulk, write pages, upload files and attach them to records.

create_record · update_record · write_records · write_page · file_upload
review

Propose

Leave a proposal with its source instead of writing directly; list open proposals.

create_proposal · list_proposals
trust

Check

Where a value came from, the fragment it was taken from, and what changed since a moment in time.

get_provenance · get_fragment · delta
rules

Rules and instructions

The rules of a space, database or record, and step-by-step instructions written by your team.

get_rules · get_instructions
schema

Understand the data

List collections and describe their fields before writing.

list_collections · describe_collection
Skills

Your team’s know-how, served to every agent

Instructions and ready-made commands are written once in Sedimenta and attached to a space, database or record. Agents get them over MCP; in Claude Code they also come as a plugin with slash commands.

Works with

Neutral to the model provider

Claude CodeCodexCursorOpenCodeVS CodeGemini CLIClaude.ai (OAuth)ChatGPT (OAuth)your own MCP client

See it on your own data

We’re working with a small number of teams that already use AI agents in their daily work.

Request early access