Skip to main content
Yes. The agent runs as a process on your machine — it is the component that executes your AI runner and processes commands. If your machine is off or the agent process has stopped, the bot will not be able to run tasks.For always-on availability, run the agent on a server, a cloud VM, or a machine that stays powered. Use a process supervisor like pm2 or systemd to restart the agent automatically after reboots or crashes.
pm2 start "remoteagent start" --name my-project --cwd /path/to/project
pm2 startup
pm2 save
Start with claude-sdk — it is the default for a reason. It has the best code understanding, the most capable tool use (file read/write, bash execution), and the lowest latency since it runs in-process with no subprocess overhead. It requires an Anthropic API key.If you want a free option with no credit card, try gemini — the Google Gemini CLI has a free tier. Note that it is currently in beta.If you need multi-model flexibility or want to use local models via Ollama, aider is the right choice. It requires Python.If you have a custom tool or internal AI gateway, use custom or openclaw.
Yes. Your API key is stored in ~/.remoteagent/config.json on your local machine and is never transmitted to RemoteAgent servers. The agent calls the AI model API (Anthropic, Google, etc.) directly from your machine. RemoteAgent only sees the prompt you sent via Telegram, not the full conversation context or your credentials.For additional security, set restrictive permissions on the config file:
chmod 600 ~/.remoteagent/config.json
Yes. Run remoteagent init in each project directory to register a new agent. Each agent gets a unique ID and appears as a separate entry in the bot and dashboard.The number of projects you can register depends on your plan:
  • Trial / Pro / Team — unlimited projects
  • Solo — up to 3 projects
When multiple agents are online, the bot asks which project to target when you send a command. You can also use /project <name> to select a specific project.
When the 14-day trial expires, your agents stop accepting commands and the bot notifies you. Your data — agents, session history, config files — is preserved in full.To reactivate, choose any paid plan from the dashboard or via the link in the bot’s trial expiry message. Your agents come back online within 30 seconds of payment confirmation, automatically.
Not currently. The relay infrastructure (web app, Redis, database) runs on RemoteAgent’s managed stack. However, the relay is designed to be transparent — it does not inspect your code or store sensitive data beyond a prompt preview. The agent package itself is open source and runs entirely on your machine.Self-hosting support may be added in the future.
The bot sends you a Telegram notification immediately when an agent disconnects. The notification includes the project name and a timestamp.To bring the agent back online, run remoteagent start in your project directory, or by name: remoteagent start "project-name".To prevent unplanned downtime, run the agent under a process supervisor that restarts it automatically. See Offline alerts for setup examples.Offline alerts are available on the Trial, Solo, Pro, and Team plans.
Update the global npm package:
npm update -g remoteagent
remoteagent --version
Then restart the agent:
remoteagent start
Updates to the agent package are backward-compatible. Your existing config files and paired agents continue to work after an update.
Only on the Team plan, which allows up to 5 Telegram users. Each team member is added by the account owner from the dashboard. All team members have equal access to all agents and session history.On Solo and Pro plans, only the account owner’s Telegram user can send commands.
Delete the agent from the dashboard. This invalidates the agent’s JWT and removes it from the database. The agent process will continue running locally but will no longer be able to receive commands or authenticate with Redis.To fully clean up, stop the agent process and remove the local config:
# Stop the agent (Ctrl+C or kill the process)
rm ~/.remoteagent/agents/{agentId}.json
RemoteAgent stores:
  • Your Telegram user ID and username (from the Telegram Login Widget)
  • Your plan and billing status (from Stripe)
  • Agent metadata: project name, machine hostname, last-seen timestamp, online/offline status
  • Session records: start time, end time, status, and the first 200 characters of your prompt
RemoteAgent does not store:
  • Your API keys
  • Your source code
  • Full prompt or response content
  • File contents processed by the AI runner
The agent is supported on Windows via WSL 2 (Windows Subsystem for Linux). Install WSL 2, then install Node.js inside the WSL environment and run remoteagent from there.Native Windows support (PowerShell/CMD) is not officially tested or guaranteed. WSL 2 is the recommended path for Windows users.