A protocol your LLM follows.
A tool that applies the result.
Zero Commit works with any chat model because it doesn’t rely on special integrations. It relies on a format.
Why it works with every model
Normally, an AI answer is a mix of explanation, snippets, and “replace this part with that.” It’s written for humans, so applying it is manual. Zero Commit’s protocol changes how the model answers. You paste it into your chat once, and the model starts replying in a strict, machine-readable format: file edits and commands, nothing else.
What the protocol tells the model
- Output only file edits and console commands before the summary. No explanations, no preamble, no recap.
- Use targeted search-and-replace edits for existing files, and full files for new ones, whichever is cheaper and safer.
- Use simple commands for deleting, moving, and renaming files.
- End with a footer listing every changed file and a total count.
- If the model needs a file it hasn’t seen, ask for it by path instead of guessing.
Fewer words from the model means less to pay for and less to go wrong.
Step by step
Add the protocol to your LLM
Paste it into ChatGPT, Claude, DeepSeek, Qwen, Grok, or whichever model you prefer.
Give it your project
Use Zero Commit’s repository ingestion to turn your whole project into one readable text file, or use Deep Map for large codebases.
Ask for what you want
Describe the feature or fix in plain language.
Paste the answer into Zero Commit
Select your GitHub repository or local project folder.
Click Detect and Stage
Zero Commit finds each operation and stages the changes.
Review, refine, ship
Use Turns to iterate, then Review & Ship to open a pull request.
Show the model your project without drowning it
Repository ingestion. Convert your repository into a single text file with your project structure and source code, so an LLM can see how your files relate.
Deep Map. When a repo is too big to paste, give the model a structured map instead. It identifies the files it needs for the task, and you send only those. Less context, more targeted answers, and potentially lower token costs.
From chat answer to staged changes
Detect and Stage reads the response, identifies the file operations, and stages them in your chosen project. It’s built for changes across many files.
The footer’s file list and count are checked against what Zero Commit actually finds, so a mismatch gets flagged before anything is finalized.
When the first answer isn’t right
Add the model’s next response as Turn 2, then Turn 3. Each builds on the last, so you can correct and refine in place. Your history gets the finished result, not every attempt.
Questions
Do I need an API key?
No. Zero Commit doesn’t call the model for you. You use the LLM you already use and paste the response in.
Which models work?
We’ve tested ChatGPT, Claude, DeepSeek, Qwen, and Grok. Any model that follows the protocol should work.
What if the model makes a mistake?
It can. Zero Commit flags mismatches between what the model says it changed and what it actually produced, and everything lands on a branch for review. It reduces risk. It doesn’t replace your judgment.
Is this an AI agent?
No. You choose what to ask, what to apply, and what to merge.
See it land your next change.
Join the waitlist for early access.