Govern AI agents in one file.
Runtime governance for AI agents in production: identity, credentials, policy enforcement, audit trail, and compliance. One file, one runtime, on your infrastructure.
Works with LangGraph, LangChain, CrewAI, OpenAI Agents, Claude Code, Cursor, Bedrock, MCP, and more.
The problem
Your agent has more access than the job needs.
It can read mail, move money, open pull requests, talk to your customers. One bad prompt and you find out the hard way.
The surface
Every tool you give an agent is a way to act on the world. Most agents today have far more tools wired up than any one task actually needs.
The failure mode
A prompt injection, a confused model step, a stolen API key. You don't know which one it'll be. You know one of them is coming.
The blast radius
When it lands, it lands in production. The agent had real credentials. Whatever it did, it really did.
How Faramesh works
One runtime. One file. One verifiable log.
Faramesh runs alongside your agent, reads your policy, decides every tool call, and writes a tamper-evident audit trail. Works with the framework you already use.
01 Runtime
Runs alongside your agent.
A daemon you launch on your own infrastructure. Every tool call goes through it first.
$ faramesh dev
02 Policy
One file. One language.
Declare what each agent can do. Versioned in your repo.
agent "payments-bot" {
rules {
permit search_docs
defer stripe/refund if amount > $500
deny! stripe/payouts
}
}03 Audit
Every decision, hash-chained.
Tamper-evident log of every decision. Verifiable offline.
04 Compatibility
Works with the framework you already use.
Framework guides for the integrations you already use. Python and TypeScript SDKs. MCP-native.
The governance file
One file defines the entire policy.
Your governance.fms is the entire policy. Every tool call goes through the faramesh daemon first, gets a decision, and lands in a hash-chained log you can verify offline.
Every tool call gets checked against the rules block. permit, defer, and deny are evaluated top to bottom. First match wins. Conditions can check the amount, the host, the method, or any argument the agent passes.
agent "payments-bot" {
default deny
rules {
permit stripe/charge if amount < $500
defer stripe/charge if amount >= $500
defer stripe/refund
deny stripe/payouts
permit stripe/customers/* method GET
}
}Quickstart
From empty repo to governed agent.
Three commands. That's the whole install.
Faramesh inspects your repo, finds the agent framework you're using, and writes a starter governance.fms. Every tool starts deferred, nothing runs until you review.
✓ Framework detected: langgraph✓ Tools discovered: 3 (search_docs, send_email, charge_card)✓ governance.fms writtenNext steps:faramesh dev run governedfaramesh apply enforceRegistry
An ecosystem of policy, not a marketplace of features.
Pre-built providers, policy packs, and framework profiles you can import like Go packages. Pinned, signed, Git-native.
Every import is pinned.
Floating tags rejected at faramesh check. Same source, same hash, every audit. Upgrade with one line change, audit with faramesh plan.
Read the versioning policySigned at every layer.
Provider binaries verified with Ed25519. Policy packs signed by the registry key. Community packs ship with their own publisher key.
See the trust modelJust a Git repo.
No separate registry service. Fork the catalog. Use it as-is. Contribute back via pull request.
Browse the catalogEvery import is pinned to exact semver. The CLI resolves them from GitHub at faramesh apply. No floating tags. No surprises in production.
Governance you can deploy in three commands.
The same enforcement at every layer. Pinned, signed, deterministic.
$ curl -fsSL https://raw.githubusercontent.com/faramesh/faramesh-core/main/install.sh | bash$ faramesh initOpen source. Self-hosted or cloud. No credit card required.
Talk to us
Two ways to get started.
Got a question?
Faramesh is open source and the community is on Slack. Ask questions, share what you’re building, and follow development.
Deploying in production?
If you’re standing it up in a regulated environment, integrating with existing identity infrastructure, or thinking about Faramesh Cloud for your team, we’d like to talk. We read everything that comes in.