Analysis updated 2026-05-18
Trace which AI agent prompt introduced a bug by clicking a suspicious commit in the graph and reading the exact prompt and diff side by side.
Audit whether any recent Claude Code or Codex session read a .env file or private key before pushing code to the repository.
Give your AI coding agent a memory of the codebase by connecting brain0 as an MCP server so it can check a file's risk history before editing it.
| brain0-ai/brain0 | proxyshard/shardbrowser | tuanle96/zca-desktop | |
|---|---|---|---|
| Stars | 22 | 22 | 22 |
| Language | Rust | Rust | Rust |
| Setup difficulty | moderate | moderate | hard |
| Complexity | 3/5 | 3/5 | 4/5 |
| Audience | ops devops | developer | developer |
Figures from each repo's GitHub metadata at analysis time.
Requires Node.js 20+, local Ollama models improve summaries but are optional, works offline without them.
Brain0 is a tool that answers a question growing more important as AI coding agents write more and more code: not just what changed in a commit, but why it changed, which AI prompt wrote it, and whether you can trust it. Git tells you what files were modified, brain0 builds a layer on top that links each commit to the specific agent session and prompt behind it. The tool works entirely by observation. It reads your git history and the session transcripts that coding agents like Claude Code and OpenAI Codex already write to disk automatically. It requires no changes to your codebase, no hooks, and no cooperation from the agents themselves. Running a single command starts the indexing process and opens a visual graph in your browser where you can explore commits, files, and functions alongside the prompts that produced them. Three specific insights that git alone cannot provide: First, drift detection compares what an AI agent said it changed in its explanation with what it actually changed in the diff. If an agent described a small update but touched twenty files, brain0 flags the gap. Second, a data-focused audit layer records which files each agent session read before writing code, flagging cases where sessions accessed environment files or files containing secrets before pushing changes. Third, a risk score tracks each change based on how many things it affects and updates later if the change was reverted or immediately patched. Brain0 also works as a memory layer for your AI agent via an MCP server connection. Your coding agent can query the graph before touching any code to understand the risk and history of what it is about to modify, including which prior prompt last changed a function and whether similar changes were later reverted. The core is written in Rust and the GUI in TypeScript. It runs offline by default using SQLite for storage and local models via Ollama for summaries, with no data sent to external services unless you configure a hosted API key. It requires Node.js 20 or later and is released under the Apache 2.0 license.
A passive observer that links every git commit to the AI agent prompt behind it, catches when agents misrepresent what they changed, and logs which sensitive files agents read before pushing code.
Mainly Rust. The stack also includes Rust, TypeScript, SQLite.
Apache 2.0: use freely for any purpose, including commercial use, as long as you keep the copyright and license notices.
Setup difficulty is rated moderate, with roughly 30min to a first successful run.
Mainly ops devops.
This repo across BitVibe Labs
Verify against the repo before relying on details.