Frequently Asked Questions

Commands.com is a secure relay for local AI agents. You run an AI agent on your machine — connected to your codebase, tools, and data — and share access with teammates, clients, or collaborators through our encrypted relay. They chat with your agent from their browser. Your code never leaves your device.

1. Install the Commands.com agent on your machine
2. Point it at a local directory (your codebase, project files, etc.)
3. Choose a permission profile (read-only, dev-safe, or full)
4. Share the session link with whoever needs access
5. They chat with your agent from their browser — you monitor everything locally

See our Getting Started guide for detailed setup instructions.

• Share a codebase: Let a product owner, manager, or teammate ask questions about your local code without giving them repo access
• Remote collaboration: Give a contractor or colleague access to explore your project through an AI agent
• Tech support: Help someone remotely — share a scoped agent that can read their files and suggest fixes
• Access from anywhere: Start an agent at your desk, continue the conversation from your phone or another device

No. Your code stays on your machine at all times. The agent runs locally and communicates through our encrypted relay. Commands.com never sees, stores, or processes your code. We relay encrypted bytes — nothing more.

Yes. The platform is 100% free and always will be. The gateway is open source — you can self-host it on your own infrastructure with unlimited devices, unlimited messages, and zero cost.

We also offer a hosted relay at api.commands.com for teams who prefer managed infrastructure. The hosted relay has a free tier and optional paid plans. See our Pricing page for details.

Every session uses end-to-end encryption with modern cryptographic primitives:

• Ed25519 for agent identity and authentication
• X25519 ECDH for per-session ephemeral key exchange
• HKDF-SHA256 for key derivation
• AES-256-GCM for message encryption with replay protection

Session keys are derived independently by the agent and the browser. Commands.com's relay only verifies the agent's signature — it never holds encryption keys and cannot decrypt any messages.

No. This isn't a policy decision — it's an architectural one. The relay passes encrypted bytes between your agent and the remote user. We don't hold the encryption keys, so we can't decrypt the messages even if we wanted to, even if compelled to. Your API keys, credentials, and code never leave your machine.

Permission profiles control exactly what the agent is allowed to do on your machine. You choose the level when you start the agent:

• Read-only: The agent can read files, search code, and list directories — but can't modify anything or run commands. Ideal for letting someone explore your codebase.
• Dev-safe: Adds the ability to run safe commands and use additional tools, with blocked paths (like .ssh, .aws) and restricted bash patterns. Good for trusted collaborators.
• Full: Unrestricted access to all tools. Only use with people you fully trust.

You can also create custom policies with fine-grained control over allowed directories, blocked paths, and specific tool permissions.

Yes. Every prompt and response is logged locally in an append-only audit trail on your machine. You can monitor sessions in real time and terminate any session instantly. The audit log records who sent each message, what was asked, and whether encryption was active — giving you full visibility into how your agent is being used.

The remote user opens a session link in their browser. They see a simple chat interface where they can ask the agent questions. No installation required on their end — they just need a browser. The agent processes their questions locally on your machine and sends encrypted responses back through the relay.

Yes. You can terminate any session instantly from your local machine. You can also simply stop the agent process — once it disconnects from the relay, no further communication is possible. Session encryption keys are ephemeral and exist only in memory, so they're destroyed when the session ends.

Yes, the agent must be running on your machine for remote users to interact with it. If the agent disconnects (machine sleeps, network drops, etc.), it will automatically reconnect when connectivity resumes. The agent uses minimal resources — roughly 50MB of RAM and negligible CPU when idle.

No. The agent makes an outbound WebSocket connection to Commands.com's relay. No inbound ports, no firewall changes, no port forwarding required. It works through corporate firewalls, NATs, and proxies — anywhere you can make an HTTPS connection.

Yes. Common team use cases include:

• Product owners querying developers' codebases without needing repo access
• Senior engineers sharing knowledge through an agent scoped to a project
• Onboarding new team members with read-only agent access to the codebase
• Cross-team collaboration where each team shares their own scoped agents

Teams can self-host the open source gateway with no limits, or use our hosted relay for managed infrastructure.