A calm workspace for AI coding agents.
CodeRight is a native desktop app for supervising AI coding agents — not another IDE. Bring any model. Watch the work. Approve what matters. Local-first, and light enough to forget it's running.
Native .NET + Avalonia — no Electron, no bundled browser, no cloud account.
+ return returnTo || "/dashboard";
What it is
A command deck for your agents — not another editor.
You don't write in CodeRight. You run agents from it: start a session, point it at a project, pick a model, and supervise. The work is visible, the dangerous steps ask first, and the room stays quiet until something actually needs you.
Model-agnostic
Bring any model over an OpenAI-compatible endpoint — Kimi, DeepSeek, GLM, Qwen, Mistral, or anything on OpenRouter, NVIDIA NIM, Groq, or Cerebras. No vendor baked in.
Local-first & private
Sessions, transcripts, and API keys live on your machine, encrypted at rest. No account to create, no telemetry, no copy of your code on someone else's server.
Native & light
Built on .NET + Avalonia — no Electron, no bundled Chromium. A small resident footprint in tens of megabytes, not gigabytes. It gets out of your way.
Built to be trusted
Reproduce-before-fix, verify-before-ship, and explicit approval gates are part of the runtime. One copper dot means an agent needs you — nothing blinks for attention it hasn't earned.
Models
Your keys. Your models. Your call.
CodeRight talks to providers over a standard OpenAI-compatible transport, so you switch models per session without switching tools. Use a fast cheap model to draft and a stronger one to review — in the same window.
How it works
Calm by design.
Supervise, don't babysit
Each project holds its sessions, each session runs its own agent with its own model and permissions. Glance at the sidebar; the status dots tell you who's working and who's waiting.
One attention color
Copper means one thing: an agent needs your decision. Sage means verified. Everything else stays quiet. Your attention is the scarce resource the whole interface protects.
Dangerous steps ask first
Edits, shell commands, and anything that ships pause for an explicit approval — with the exact change in front of you. Trusted work flows; risky work waits. You set the line.
Evidence before "done"
Agents reproduce a bug before fixing it and verify before claiming success. Approvals are bound to the exact content they reviewed — change the file and the green seal goes stale.
Waitlist
Get the beta when it's ready.
CodeRight is in active development. Drop your email and I'll send a download link when the Windows beta is ready — nothing else.