Local control plane for AI agents
Run OpenClaw, NanoClaw, and future agents with local sandboxes, guided setup, credential handling, guardrails, approvals, jobs, logs, and runtime health in one place.
Run any agent. Securely. Local-first by default.
Experimental release candidate: Armorer is under active development. The current release train is intended for early testers who are comfortable with local agent runtimes, Docker/Colima, and rapidly evolving setup flows.
Website · Install · Docs for humans · Issues
Star ArmorerLabs/Armorer if you want safer local agent runtimes to exist.
One Command
curl -fsSL https://armorerlabs.com/install | shThen open the local UI and install supported agents through Armorer.
The default installer uses the native armorer launcher and a pinned local
appliance image. Users do not need Python, Node.js, pnpm, Rust, or a source
checkout.
armorer status armorer doctor armorer logs --follow
Develop
curl -fsSL https://armorerlabs.com/install | sh -s -- --devOr run the workspace manually:
corepack enable pnpm install --frozen-lockfile pnpm build pnpm test pnpm typecheck pnpm --filter armorer-ui-selfhost build
Source/development runs also need Rust/Cargo for the native Armorer gateway.
pnpm dev, pnpm start, and pnpm build run a gateway preflight that uses
the packaged gateway when available or builds crates/armorer-gateway locally.
Start the self-hosted UI from source:
pnpm dev -- --hostname 0.0.0.0 --port 3099
Then open http://127.0.0.1:3099.
Use the CLI directly from the workspace:
pnpm armorer -- list
pnpm armorer -- core status
pnpm armorer -- install nanoclaw --yes --skip-scan
pnpm armorer -- jobs listUse Your Agent
Set up Armorer from https://github.com/ArmorerLabs/Armorer on this machine.
Follow AGENTS.md and the repository instructions.
Install Armorer, verify Docker, start the local UI, then help me install and configure OpenClaw through Armorer.
Do not report success until the Armorer CLI works, the UI is reachable, and runtime health checks pass.
Why Armorer
| Running agents directly | Running agents through Armorer Labs |
|---|---|
| Each agent has its own setup path | One local control plane |
| Credentials are easy to scatter | Guided credential handling |
| Risky actions are hard to review | Human approvals and guardrails |
| Logs live wherever the agent puts them | Jobs, runtime status, and audit trails |
| Local networking gets messy | Local-first with optional private remote access |
What Exists Today
- Local agent runtime and UI.
- Docker sandbox orchestration for installed agents.
- Guided setup for providers, models, credentials, channels, and skills.
- OpenClaw and NanoClaw as first-class managed agents.
- Armorer Agent chat runs with durable Flue-backed run state and live workstream visibility.
- Guardrails for suspicious inputs, leaked credentials, and risky actions.
- Jobs, logs, runtime health, and setup state in one operator surface.
Humans
Prefer doing it yourself instead of handing the repo to an agent?
Start here: HUMANS.md
Positioning
Armorer Labs is not another model provider. It is not a hosted agent platform. It is a self-hosted, local-first agent runtime layer for builders who want control, sandboxing, and security posture before they scale agent usage.