GitHub radar
A coding agent harness that never hides or trims anything
UNCUT is a free, open-source program that runs an AI coding agent and shows you exactly what you sent it and what it sent back, with nothing cut.
It's a small go-between program that passes your whole conversation to an AI model, never shortening history, files, or command output along the way.
When it helps
Useful for scripts, automated test runs, and isolated containers where you need full predictability. Skip it if you want a chat window or built-in top-tier answers — quality depends entirely on the model you attach.
Pros
- Never trims conversations, files, or command output
- Only 7 MB, runs on any laptop without a GPU
- Just 837 lines of code — easy to read and fork
- Works with any paid API model or local Ollama
Cons
- No sandbox — the model runs shell commands as you
- No chat window or auto-retries — you're on your own
- Answer quality depends entirely on the model you attach
How to set it up — step by step
- 1Open your AI agent (Claude Code, Codex) inside your project folder.
- 2Send the agent the text from the block below.
- 3Give the agent a paid provider's API key, or ask it to set up a local Ollama.
- 4Check that the uncut --version command responds after install.
- 5Run uncut on a git diff in one of your repos and read its answer.
Text for your agent
Copy this and send it to your agent — Claude Code, Codex, any of them:
Set up UNCUT from here: https://github.com/Balanov/uncut — follow the README (download the prebuilt binary for my OS or build it from source with go build), connect a model through environment variables (ask me for an API key if a paid provider is needed, or set up a local Ollama model instead), and run it on git diff in one of my repos — show me what it answers.
The tool itself is tiny — a 7 MB binary that runs on any regular laptop, no GPU needed. But it always needs a model behind it: either a paid API key from a cloud provider, or your own model through Ollama — then hardware depends on that model's own requirements.
Open on GitHub▌ More finds