Sign in Start free

[The Agent OS]

Put agents to work.

Every run under identity, policy, audit, budget, and compounding skills.

Start free Have us run it for you

The governed boundary

Everything an agent can do —and the one gate it all crosses.

Toggle eleven built-in capabilities on. Every call the agent makes is checked against policy, paused for you the moment it's irreversible, and written to the journal. That boundary is the platform — not a feature you bolt on.

noorle/agent-os booting kernel…

capability manifest · agent-04

agent-04 durable process · learned skills 7 running
  • web_searchsearch the web
  • memorydurable facts across threads
  • browserheadless render + extract
  • http_clientany http api
  • knowledgerag over your docs
  • filesworkspace read + write
  • code_runnerrun python and js
  • sandboxdisposable linux
  • computerpersistent vmalways-on
  • plugin_builderbuild new tools
  • skill_builderauthor skills

+ connectors · plugins (wasm) — the same shelf, extended

autonomy gate

journal · audit trace

    metered 0 tokens · 00:00 runtime · $0.000 world-cost tokens, time, world-cost — nothing else
    The platform

    Three surfaces. One substrate.

    01 · agents Autonomous agents Every call governed and written to the journal; skills that compound each run — on a durable runtime that survives deploys and reroutes around model failures. Agents → 02 · gateways Composed gateways Publish your capabilities over MCP — to your team's AI tools, your clients', or the registry, under your name. Gateways → 03 · workflows Durable workflows A Program run once, or published as a Workflow — waits, approvals, and Automations to start it. Workflows →

    The substrate is capabilities — built-in tools, your sandboxed code, connectors — bound once, governed at the source, consumed by all three surfaces.

    Frameworks help agents think. An Agent OS governs what agents can do.
    the position, in one sentence
    What “governed” means here

    An OS earns its name by governing what runs on it.

    Every effect an agent has on the world crosses a gate it can't skip — checked on every call, written to the journal, and enforced by the platform, not by whoever wrote the agent.

    01 · identity Identity Every capability call carries an account and an actor.
    02 · capabilities Capabilities Granted and scoped — never ambient access.
    03 · policy Policy Allow, deny, or pause for a human. Decisions freeze on the record.
    04 · isolation Isolation WASM sandboxes, allow-listed hosts, brokered VMs.
    05 · durability Durability Runs and memory survive restarts and deploys.
    06 · audit Audit A per-call journal with cost attribution, forever.
    Where Noorle sits

    Not a framework. Not a firm. An Agent OS.

    Build it yourself

    Open-source harnesses hand you a runtime. You author every role, hold the keys, and you're the one enforcing trust.

    Have it done

    Forward-deployed teams deliver trust as an engagement — scoped as a project, not a product.

    Noorle

    Trust and judgment as product: a gate nothing skips, an audit journal by default, skills that compound.

    The engineering journal

    Latest from the log.

    Cross-section of a multilayer circuit board, a single copper via fanning into three parallel traces across distinct strata.

    One Source, Three Durability Tiers: How Noorle's Program Runtime Is Built

    Durability isn't one mechanism — it's a spectrum. A well-built orchestration runtime picks a different point on it for each kind of state it owns, on purpose.

    Read →     Abstract circuit board pattern representing the boundary between an agent's control plane and its execution environment.

    Outside the Sandbox: Why Production Agent Runtimes Need a Hybrid Architecture

    Most agent platforms force you to choose between security and state — orchestration inside the sandbox is convenient but unsafe; orchestration outside with ephemeral sandboxes is safe but stateless. The hybrid pattern resolves the trade-off.

    Read →     Illustration of the shift from AI agent frameworks to agent runtimes.

    Frameworks vs. Runtimes: The Next Shift in AI Agent Infrastructure

    Frameworks taught us how to build agents. Runtimes are where agents actually run. Understanding this distinction is the key to shipping agents that work in production.

    Read →
    Two ways in

    Build on it — or have us run it for you.

    The wide door · self-serve

    Build on the platform

    Compose capabilities and ship a governed agent this week — for your own operations, or for clients you serve. Start with a small job or from scratch.

    Start free See pricing
    The narrow door · done for you

    Works runs it for you

    Your back office still runs by hand? Our own team builds and runs an agent workforce for your business — starting with a pilot mission.

    Request a pilot mission → limited client slots