What it does
OpenAgentRelay is a small command-line tool for sharing something that already works on one computer with other people or agents. The publisher wraps an existing command with relay serve, such as a script or a restricted Codex or Claude Code invocation. Callers install only the relay CLI and use relay ask with the publisher's URL and an access key, so they never need the publisher's source code, dependencies, prompts or credentials. Relay starts the command once per request, passes the request on standard input and returns the final answer from standard output. Callers can ask for JSON output, and a nonzero exit code signals failure, which suits other agents and automation. Follow-up questions can use a relay-managed conversation that is kept in memory and expires after one hour by default. The main branch is labelled an Alpha for direct calls on a trusted LAN. It uses plain HTTP and one shared key, and version 0.1 has no TLS, per-user permissions, automatic discovery or sandbox. The README says not to expose it to the public internet or connect it to production write operations. It installs with pipx install open-agent-relay and is licensed under Apache-2.0.
Capabilities
- relay serve publishes any command that reads a request on standard input and answers on standard output
- relay ask calls a shared capability, with JSON output and nonzero exit codes on failure
- Shared access key supplied through an environment variable
- Optional relay-managed conversations for follow-up questions
- Controls for execution timeout, request size, output size and concurrency
- Examples for a local CSV report, a Codex review and a Claude Code review
When to use it
- Letting a teammate call an agent or script on your machine without sharing its code or credentials
- Giving another agent a callable capability that returns JSON
- Trying a restricted, read-only Codex or Claude Code review on a trusted local network
