Hey Product Hunt

I built Relayvora after hitting the same wall over and over: Claude Code would hit its usage limit in the middle of a fix, and switching to Codex meant re-explaining everything. What I was doing, what I'd already tried, which test was still failing. Pasting the whole transcript into the next agent wasted tokens and still missed things.

Relayvora fixes that with one command:

relayvora handoff --from claude --to codex --verify auto

It reads the session on your machine, collects what actually happened in Git (changed files, diffs, commits from the session), runs your tests, and writes a short handoff of about 1k tokens. The next agent continues from there. Every line is labelled by where it came from: [observed] means Relayvora saw it in Git or ran it itself, while [inferred] and [agent-reported] mean the next agent should check it. That way the new agent doesn't take a claim like "tests pass" on faith.

What else it does:

• Works with Claude Code, Codex and OpenCode, plus any MCP-capable agent through the built-in MCP server

• Hooks can save a checkpoint automatically when an agent hits its limit, compacts or ends a session

• Send a handoff to a teammate with relayvora send. It travels through a branch in your own Git repo: no server and no new account.

• Local-only, no telemetry, free and open source (MIT). Windows, macOS and Linux.

Try it: npm install -g relayvora, then run relayvora in a project.

It's still alpha, so I'd really value feedback on:

1. Which agent pairs you switch between most, and which one I should support next

2. Whether the handoff gives the next agent enough context, or too much

3. Anything that broke or felt confusing on first run

I'll be here all day answering questions. Thanks for checking it out!