What is Recoder?

Recoder is a desktop agent control plane. The product is a terminal: you open Recoder, run your chosen coding agents in parallel isolated git worktrees, then review and merge their work — with governance and a full audit trail underneath.

The problem it solves

A single AI coding agent gives you one attempt at a task. If you want a second opinion you run it again and lose the first. If you run several agents at once, you're left with N divergent branches and no principled way to combine them.

Recoder turns N divergent parallel runs into one trusted merge:

• •Parallel, isolated — each agent works in its own git worktree; they never step on each other.
• •Compare — see where agents agree, where they diverge, and which files are contested.
• •Converge — per file, take one agent's version wholesale (pick) or merge non-overlapping edits from several (synthesize).
• •Govern — every merge passes a policy gate, can require human approval, and is sealed into a tamper-evident audit chain with signed evidence you can export for SOC 2 / EU AI Act / DORA reviews.

Who it's for

• •Individual developers who want to run a few agents against a task and keep the best result, fast.
• •Teams who need parallel agents and a record of what happened — who changed what, which policy allowed it, and proof it was reviewed.
• •Regulated / enterprise environments that need agent work to be governed and auditable, self-hosted, with no vendor cloud in the loop.

What it is not

It is not an editor — Recoder doesn't replace your IDE or add tab-completion; it drives the agent CLIs you already run. And it is not a token reseller — you bring your own provider keys; Recoder orchestrates and never resells inference.