What's the first thing you'd let an agent do on your own machine?

[prax211]

We just open-sourced agentic-remote-pc — one authenticated gateway (REST + MCP) that turns any Windows or Linux PC into a secure, agent-controllable workstation. ChatGPT, Claude, Cursor, Codex, and Gemini can run real shells and drive 11 coding-agent CLIs on a host you control, behind your own key and your own tunnel.

A few things it enables:

• Real shells (PowerShell/CMD/bash/zsh) + agent CLIs (Claude Code, Cursor, Aider, OpenCode, Gemini CLI, Codex CLI, Copilot CLI, Goose, Amp, Qwen, Crush)
• Agent-to-agent orchestration on real hardware — a cloud model drives a local Claude to implement, then a local Cursor to validate
• LiteLLM gateway wiring so a local/open model (Ollama/vLLM/LM Studio) gets tool access to a real host
• Cloudflare Tunnel or a fully self-hosted relay (frp/rathole/bore/chisel) — no vendor lock-in

Security stance (stated plainly in the README): the default single-bearer-over-TLS config is for personal/dev/QA. Production needs OAuth 2.0 (Cloudflare Access or an OAuth2 reverse proxy), TLS, least-privilege accounts, and command sandboxing.

Repo: https://github.com/cogniziocompany/agentic-remote-pc
Blog: https://cognizio.company/blog/agentic-remote-pc.html

We'd love your feedback. A few questions to kick things off:

1. What's the first workflow you'd hand to an agent on your own machine?
2. Which agent CLI or shell would you want supported next?
3. For production deployments — what identity/access pattern are you running (Cloudflare Access, Tailscale, an OAuth2 reverse proxy)?

Drop your use cases, friction, and feature requests below. We'll track the top asks and follow up.