RemoteAgent runs on the operator's own machine, executes under least-privilege policy, and proves every step with a tamper-evident audit log. The brain lives in the cloud; the hands — and the policy — stay local.
Remote work is a trust problem. RemoteAgent turns "trust me, I did it" into "verify it yourself."
Describe the work and attach a policy profile — which capabilities the worker's agent may use, which workspace it may touch, and what needs approval.
employer sideThe worker's RemoteAgent runs the job on their own machine. Least-privilege capability grants, workspace-confined files, and an argv-based shell enforce the rules — locally.
worker sideEvery action is appended to a hash-chained audit log. The worker exports an evidence bundle; the employer verifies it before releasing escrow.
verified, not promisedVerify any run yourself: remoteagent audit verify
Employers and workers flow into the same client, the same policy engine, and the same audit chain.
Post work, specify the policy, and review proof-of-work instead of trusting a portfolio claim. Escrow holds the payment until the audit bundle checks out.
Install the open-source agent on your own machine, set your policy, and claim gigs. The agent does the mechanical, verifiable work — you review, approve, and deliver proof.
Everything below is the core product — not a bolt-on. That's why the funnel strengthens the main SaaS instead of diluting it.
The device enforces its own policy. You set the rules; the machine keeps them, whether or not the network agrees.
Least-privilege by default. Shell, files, and network are granted explicitly — and grants narrow, never union.
Every action is hash-chained into a log you can verify with remoteagent audit verify and export as evidence.
File access stays inside declared workspace roots. The agent can't wander into the rest of the machine.
Commands run as argument arrays — no shell string interpolation, no injection through a stray space or quote.
Outbound fetches are validated per-hop. Internal addresses are refused by default, not by accident.
Trusted instructions and untrusted content are kept apart, so a job description can't quietly rewrite the plan.
Task routing, model orchestration, and the dashboard live at api.remoteagent.online. Execution stays on your device.
The runtime is MIT and free. The control plane is the commercial product — the same one the main site sells.
In remote work, the scarce thing is evidence. RemoteAgent produces it automatically: each step is recorded and chained, so the employer can verify exactly what ran, on which device, under which policy — before any payment is released.
Start free on either side. The client stays open source and free; you pay for the cloud brain and the marketplace.
Commission: 10% platform fee (7% on Pro) + optional 3–5% employer escrow/processing fee. Metered agent-hours sold in blocks above the included allowance. Job boards earn 25% rev-share or a flat CPA. Marketplace tiers stack on your RemoteAgent plan (Free · Pro $12 · Team $25).
No. The agent runs on the operator's own machine — the worker's laptop, desktop, or server. Only the "brain" (task routing, model orchestration, the dashboard, and audit storage) lives in the control plane at api.remoteagent.online. The device dials out over WebSocket; nothing dials in.
Every action is appended to a hash-chained, tamper-evident audit log. A run can be exported as an evidence bundle and checked with remoteagent audit verify. If anyone altered the log, the chain breaks and verification fails.
The device enforces its own policy. Capability grants are least-privilege, file access is workspace-confined, the shell is argv-based (no string interpolation), and network access is SSRF-safe. The operator sets the policy; an employer can require a policy profile for a given job.
No. A human operates RemoteAgent as a tool on their own machine and remains accountable for the outcome. The audit chain makes the work inspectable rather than a black box. Approval gates keep a person in the loop for sensitive steps.
The heavy reasoning — understanding the task, planning steps, routing to a model — happens in the control plane. The agent on the device carries out the approved steps under local policy. The plan is coordinated remotely; the hands are local and policed.
Work that is verifiable and runs on a machine: writing and testing code, running data pipelines, generating reports, drafting content, and ops/reliability tasks. It's not for physical work, and not for anything where a tamper-evident log can't capture the outcome.
Yes — the runtime is MIT-licensed and free. The control plane (the commercial product) is what you pay for. This is the same model documented on the main RemoteAgent site.
Install the client (curl -fsSL https://remoteagent.online/install.sh | sh), sign in with remoteagent login, start the daemon with remoteagent start, set your policy, and connect. Employers sign up at app.remoteagent.online and post a job with a policy profile.
RemoteAgent is also a self-hosted AI agent runtime for security teams — local policy, capability grants, and a tamper-evident audit log. This jobs page is one door into the same product.
Visit remoteagent.online →