Remote control Claude Code, Codex and DeepSeek harness (CLI) from phone
Also works with Pi agent — and with a custom model or your own API endpoint, where official remote control is turned off.
Your session keeps running on your own machine — reconnect from phone, browser or desktop.
Hand a live connection to family or a teammate — they drive your agent without your password. Revocable, with an expiry you choose.
Agents deployed on different machines join one encrypted network automatically — they discover each other and talk directly. No public IP, no port forwarding.
How remote Claude Code access works
Your coding agent keeps running on your own machine — your files, your git checkout, your environment. PandaNpc runs a small bridge next to it and gives you a live connection from anywhere, so you can send a prompt, approve a tool call, or read the diff without being at that desk.
Reach it from an iPhone, any browser, or the desktop app on Windows, macOS and Linux. Sessions survive the network dropping — reconnect and the conversation is still there, still running.
Works with the agent you already run
Claude Code — Drive a live session from your phone: send prompts, answer permission prompts, watch subagents, read the diff. Your terminal keeps the same conversation, so you can hand off mid-task in either direction.
Codex CLI — The same remote session, approvals and history for OpenAI Codex CLI running on your own machine.
DeepSeek harness (CLI) — Connect to the DeepSeek harness the same way, with no separate setup on the phone side.
Pi agent — Also supported, over the same connection.
Works when official remote control doesn't
Anthropic's own Remote Control requires your session to talk directly to api.anthropic.com on a Pro, Max, Team or Enterprise plan. Its documentation lists what turns it off: API-key authentication, Amazon Bedrock, Google Cloud's Agent Platform, Microsoft Foundry, and any custom ANTHROPIC_BASE_URL pointing at an LLM gateway or proxy — even if you are signed in with claude.ai.
PandaNpc talks to the CLI on your machine, not to a model provider. Whichever endpoint, gateway or model you have configured, the connection works the same — and your code never leaves your machine.