brain0

brain0 is an open-source, offline-first tool that creates a passive decision graph linking every commit to the AI agent prompts, context, and decisions behind it. It provides drift detection, DLP audit of what agents read, evidence-driven risk assessment, MCP memory for coding agents, and signed provenance attestations. Designed for teams using AI coding assistants (like GitHub Copilot or autonomous agents), brain0 helps maintain transparency, auditability, and security in AI-generated code. Unlike traditional version control that only shows what changed, brain0 explains why: which prompts, which models, which context led to each code change, all with one command and running offline by default. Key features include drift detection that reconciles declared vs. actual changes, a DLP layer recording sensitive file reads, and a two-dimensional risk score. It auto-discovers sessions from coding agents like Codex and Claude Code, and offers local LLM support via Ollama for summarization and semantic search. The graph explorer GUI allows exploring codebase structure from repo to symbol level, with clickable commits revealing prompts, diffs, and risk. brain0 is dogfooded by its own development team. Its offline-first, open-source (Apache-2.0) approach makes it trustworthy and flexible. Compared to cloud-hosted alternatives, brain0 ensures data stays local and gives full control over model choices. However, it requires Node.js ≥20 and optionally Ollama for enhanced features, and is passive by design — not real-time.