RemoteAgent is an AI agent runtime that lives on your computers. It dials out to the remoteagent.online control plane, executes tasks under the device's own policy, and writes every action to a tamper-evident audit log. The brain is in the cloud. The authority is on your machine.
Most AI agents ask you to hand over a shell and trust them. RemoteAgent inverts the relationship: the device grants capabilities, the policy engine enforces them, and the audit chain proves what happened.
Per-tool allow/deny/prompt rules live in ~/.remoteagent/policy.json — the device's authority. The cloud can never widen it. Capability grants intersect; they never union.
Commands are parsed into argv, binaries resolved to realpaths and checked against an allowlist. No curl | sh, no redirects, no globs. Workspace-confined file access.
Every step is hash-chained and anchored to the cloud. remoteagent audit verify detects tampering — proof you can export to a SIEM or hand to an auditor.
A real state machine: created → planned → running → verifying → succeeded. A side effect is never replayed merely because a process restarted.
Trusted instructions and untrusted tool content never share a context channel. Attack tests live in the repo and run on every commit.
Facts and preferences persist with provenance and revocation — and the policy decides what the agent may remember.
The client dials out. Nothing dials in. Your machine keeps the keys, the policy, and the audit chain.
The remoteagent daemon: executes tools, enforces policy, keeps the tamper-evident audit chain, and reports every step to the dashboard.
The control plane: task routing, scheduling, the reasoning brain, dashboards, fleet administration, upgrades, and compliance exports.
One installer script, one command, upgrades with checksum verification and dual-named archives for in-place updates from older versions.
The client is free and MIT. You pay for what the control plane does: scheduling, audit, memory, policy profiles, fleet admin, compliance exports — regardless of whose keys fire the model.
RemoteAgent powers remote jobs online: post a job with a policy profile, a worker's agent executes it under the intersection of local and job policy, and the tamper-evident audit log proves what was done — before escrow releases.
Remote jobs →On your machine. macOS, Linux, Windows, WSL2 or Raspberry Pi. The cloud is the brain and the dashboard; the device is the executor and the enforcer.
Your machine keeps running what policy allows. The daemon reconnects with backoff; local jobs, policies and the audit chain are unaffected by a network outage.
Only what your policy permits. The capability model intersects cloud requests with device policy — a grant can narrow, never widen. A denied tool is denied, logged, and visible in the dashboard.
remoteagent audit verify walks the hash chain. Export an evidence bundle for a SIEM, an auditor, or a job escrow release.
Yes — MIT, one runtime dependency, no telemetry you can't turn off. The open core is the trust story; the control plane is the product. Business logic lives server-side and never ships in the client.
Pro and Team plans: yes — provider.json or RA_PROVIDER_* env vars, on-device reasoning, zero prompts leave the machine. The plane keeps coordinating and auditing either way.
Downloads, verifies checksums, installs remoteagent (with the legacy mona-agent alias), migrates state, and starts the daemon.
Windows: irm https://remoteagent.online/install.ps1 | iex · Docker: docker run -d remoteagent/agent