Aarmos
Confidence for Autonomous Computing.
A world where every AI-executed business decision is governed, provable, and continuously improved.
"Own the agent. Govern the execution. Prove it, forever."
Decision Governance
Every AI-executed business decision is named, bound to policy, and evaluated before anything executes.
Operational Intelligence
Replay, counterfactuals, contribution, drift and override analysis — derived only from evidence already in receipts. Recomputable, attributable, and free of synthetic insight.
Continuous Improvement
Replay decisions exactly as they ran, simulate policy changes, and close the loop from proof to better governance.
The name
The name began as an internal working title: AI Action Runtime Management Operating System. As the platform evolved, it outgrew that acronym. Today, Aarmos stands for something simpler — the Governance Runtime for autonomous software: govern AI before it acts, run it within your existing stack, and prove every decision.
How it works
Aarmos evaluates every tool call against policy you author, in the hot path, with predictable microsecond-scale tail latency. Every governed action produces one signed, tamper-evident receipt that any auditor can verify offline — the receipt format is public (CC BY 4.0) and the verifier ships as a portable Apache-2.0 artifact, so your proof outlives Aarmos the company.
What we don't do
We don't read your prompts, tool traffic, model keys, or receipt evidence — enforcement and evidence stay on your device or your infrastructure. We don't ship a frontier model, a coding IDE, or a consumer assistant.
Workspace features are the exception, and they're explicit: accounts, membership, entitlements, and workspace audit event envelopes (actor, action, resource, timestamp) are stored to run Team and Enterprise governance. Payloads are never part of that. Privacy notes →
The Governed AI Manifesto
AI won't replace software. AI becomes software.
Software earned trust through version control, testing, CI/CD, observability, and security.
AI needs the same.
Aarmos exists to answer the four questions that matter when something goes wrong — or when an auditor asks months later:
- Why was this action allowed?
- Which policy approved it?
- Can we reproduce the decision?
- Can anyone verify the evidence years from now, without us?
Product principles
What the runtime enforces. For how we build and write, see the design principles below.
- Govern before execution. Decisions are made in the hot path, not reconstructed from logs.
- Vendor neutrality. Aarmos governs AI you didn't buy from Aarmos.
- Portable evidence. Proof outlives the vendor that produced it.
- Local-first. Enforcement runs on your device or your infrastructure by default.
- Open verification. The spec is public; the verifier is a portable Apache-2.0 artifact.
- Least privilege. Agents get the smallest scope that lets them work.
- Human accountability. Every decision traces to a policy, an authority, and a person.
Built for developers, teams, and security-conscious enterprises who want AI power without handing over keys, chats, or tool traffic.
Aarmos is a product of Aarmatix LLC.