Inspect

Inspect is an open-source CLI tool for Git that surfaces structurally risky code changes in milliseconds. It analyzes each function, class, or method in a diff, scores risk based on the dependency graph—dependents and blast radius—and outputs a prioritized list of entities needing human attention. Built by Ataraxy Labs, Inspect runs entirely local: no API keys, no database, no network calls for core triage. It supports 21 programming languages plus 5 data formats via tree-sitter and sem-core. Key features include entity-level diff analysis with graph-centric risk scoring, change classification using the ConGra taxonomy (text, syntax, functional), and tangled-commit untangling via Union-Find on dependency edges. An optional LLM review mode (inspect review) works with Anthropic, OpenAI, Ollama, or any OpenAI-compatible server, sending only critical entities for deeper analysis. The tool also offers an MCP server with 6 tools for AI-agent integration (inspect_triage, inspect_entity, etc.), and outputs in terminal, JSON, or markdown formats. Performance is a highlight: Inspect achieves 95% recall on planted bugs and 83.5% high/critical recall on the Greptile benchmark—beating competitors like Augment, Greptile, CodeRabbit, Cursor, and Copilot—all with zero API calls and millisecond-per-commit speeds. A recent rewrite prioritized graph signals over classification, reducing false positives by half. The four-phase pipeline (Extract, Classify, Score, Group) processes large codebases efficiently, with parallel graph building cutting Sentry's analysis from 40s to 4s. Compared to LLM-first tools such as CodeRabbit or Greptile, Inspect's core signal is structural rather than semantic—ideal for teams that want deterministic, fast, local triage without recurring API costs. It excels in CI pipelines where every millisecond counts and where developers need immediate, trustworthy risk signals. For teams that prefer a hosted UI or conversational review flow, Inspect's CLI-only