Shared across agents & people
Ad-hoc CLAUDE.mdEach person copies their own file; it drifts
FregeOne versioned memory, with authorized views by role
The governed operating layer for teams running multiple AI agents
Many agents. One organizational reality.
Frege gives MCP-connected agents one governed organizational memory: scoped, cited context and reviewable updates—without locking company knowledge to one model, account, or repository.
Shared live agent sessions (private beta, Codex) — Authorized teammates watch a running agent session live: tool calls, file changes, and approval requests as they happen. One controller can redirect the run, stop it, resolve approvals, or hand control to a teammate who continues the same thread. Execution stays on your machine; every control decision is recorded in the governed session history.
- Interface
- MCP + REST
- Control
- Organization + role
- Memory
- Cited + reviewable
Available now / governed context
The agent sees what it can use—and the source behind it.
Frege resolves the caller before assembling context. Allowed sources arrive with citations. Restricted matches stay outside the packet. Durable changes enter a human review queue instead of silently rewriting memory.
Sanitized demonstrationCurrent capability
agent → build_context("EU enterprise refund")
frege → resolve organization · key owner · role · trust zone
allowed → 3 cited sources returned
refund-policy.md@v7 · eu-addendum.md@v3 · escalation.md@v5
withheld → 2 restricted matches omitted from the packet
proposal → finance sign-off note queued for human review
actor: support-agentzone: greensources: versionedwrite: proposed
Run the complete 90-second product proof Why not just CLAUDE.md?
CLAUDE.md is great for one person on one repository.
If that is you, use it—it is free and it works. Frege is for the next step: when several agents and people depend on the same knowledge, and not everyone should see everything.
Who can see what
Ad-hoc CLAUDE.mdEverything in the file is visible to everyone
FregeRole and trust-zone gates on every request
Activity & decisions
Ad-hoc CLAUDE.mdNo shared record of the work
FregeContext builds, sessions, and access decisions recorded per org
Updating knowledge
Ad-hoc CLAUDE.mdAnyone—or any agent—edits in place
FregeAgent writes land as proposals, accepted after review
Sensitive sources
Ad-hoc CLAUDE.mdAll-or-nothing: paste it in or leave it out
FregeRestricted content stays hidden; packets report withheld counts
How it works
One controlled path from company knowledge to agent action.
- 01 / ConnectConnect Claude Code, Codex, or an internal client through MCP—or download the local Frege Agent profile for Hermes.
- 02 / Build contextFrege resolves organization, role, and trust zone, then returns scoped context with citations.
- 03 / Do workThe agent keeps its own model and tools while Frege records supported context builds and session activity.
- 04 / Improve memoryUseful discoveries land as proposals. A human accepts before canonical memory changes.
// restricted context is withheld before the packet reaches the agent
What changes for your team
Less context babysitting. Fewer wrong answers.
Frege keeps the chain from source to agent context visible and keeps proposed changes separate from accepted memory, so your agents can move faster without losing control.
- 01
Stop re-pasting context
Agents pull current, cited context on request instead of receiving the same runbook in every session.
- 02
Stop agents reading the wrong document
Scoped retrieval returns the source for the task—not a stale fork or a document the actor should not see.
- 03
Stop silent knowledge drift
Durable updates are reviewed before they become canonical, so organizational memory stays trustworthy.
Product roadmap
Start with governed memory. Build toward the operating layer.
The category is the destination; status labels show what teams can use today and what Frege is building next.
Use today
Available now
- Versioned organizational memory
- Scoped, cited context
- Reviewable memory updates
Policy first
Building next
- Service principals
- Versioned authorization policies
- Authorization receipts
Deliberate expansion
Later
- Durable tasks and workflows
- Durable workflow approval gates
- Optional hosted execution
Persistent by design
Models change. Your organization remembers.
Create your account, choose a plan, and connect the agents your team already uses—or run the downloadable Frege Agent locally. Billing activates through Stripe when you are ready.