30 minutes to confidence · on your machine

Your first governed AI agent.

One agent. One policy. One Receipt. One Replay. One offline verify. At the end you can hand a security reviewer proof they can check themselves — no sales call, no shared control plane.

Install the runtime.

~5 min

Signed artifact for macOS, Linux, or Windows — acquired from your account once your workspace is authorized. No public download.

aarmos login      # authorize this device
aarmos --version

How you get the software. Aarmos is distributed to authorized workspaces as signed artifacts. Signing in authorizes your workspace for evaluation automatically — no person to wait for — and the software is available the moment you choose to install it. Nothing downloads until you ask for it, and nothing is published to the public npm registry today. The open @avar-standard/* verifier packages are unaffected and remain publicly installable, so anyone can check a receipt without our software.

1Sign in. Your evaluation workspace and its bounded authorization are ready immediately.
2Choose. Acquisition follows need — the software is fetched when you decide to install it.
3Authenticate. Pair the CLI with your workspace from the device you will run on.
4Acquire. Pull the signed artifact at a fixed version for your authorized workspace.
5Verify. Check the artifact digest and signature before you run it.
6Run. Start the runtime and record your first governed decision.

No daemon. Nothing runs until you tell it to.

Initialize a workspace and write one policy.

~10 min

aarmos init my-first-pilot && cd my-first-pilot

Edit policy.yaml. Allow the safe path, deny the dangerous one. Start narrow:

allow:
  - tool: http.get
    when: target.host in ["api.internal.example.com"]
deny:
  - tool: shell.exec
  - tool: fs.write
    when: target.path startswith "/etc"

Run one agent through the gate.

~5 min

Point any agent at the local runtime. LangGraph, CrewAI, raw HTTP, custom — pick one. Every action now passes the policy first.

aarmos run ./agent.py

Expected output: a decision per action, each with a Receipt ID.

[allow] http.get api.internal.example.com/orders · r_01HXZ…
[deny] shell.exec — rule: no shell.exec · r_01HXZ…

Replay a decision.

~5 min

Pick any Receipt ID from the run and rerun the decision against the pinned policy. Same inputs → same verdict, same rule, same reason.

aarmos replay r_01HXZ…

Verify the Receipt offline.

~5 min

The portable verifier runs without Aarmos. Hand a Receipt to your security reviewer and they can check it on their own machine, forever.

avar verify ./receipts/r_01HXZ.json

ok · chain intact · policy pinned · signature valid

Done.

You now have a governed agent, a signed Receipt, a replayable decision, and an independent verification. Add the next agent when your team is ready.

Show your reviewer the trust posture Forward to Security Forward to Platform Full quickstart

GovernCan this agent do this?RunWill this work with my stack?ProveCan I prove what happened?

One decision. Three checkpoints. Every time.