Git Lens: AI-Powered Onboarding Tours

Problem & Challenge Onboarding new developers to a large codebase is slow and error-prone. Documentation is often outdated or missing, and commit history is a firehose of noise—most commits are chores, dependency bumps, or formatting changes. New hires spend weeks piecing together context from scattered PRs, issues, and tribal knowledge. Solution Git Lens automatically generates a personalized, narrative-driven onboarding tour directly from a Git repository. It ingests commit history, applies heuristics (reverts, incident-linked fixes, repeated-fix files, architecture-shaping changes) to filter signal from noise, then uses large language models to rank the most significant commits and narrate them into a coherent story. The result is a sequence of "stops"—each a pivotal moment in the codebase's evolution, complete with title, narrative, subsystem, and supporting documents (e.g., linked GitHub issues or Jira tickets). How It Works Extraction – Mines the repo's history, partitions it into subsystems, and flags candidate commits using configurable heuristics. Ranking – A multi-stage pipeline summarizes files and commits, builds a project overview, and ranks candidates using an LLM, with a deterministic post-processing step to ensure diversity and quality. Narration – For each selected stop, it fetches relevant diffs and linked documents, then generates a 2–4 paragraph narrative tailored for a new hire. Caching – Aggressively caches intermediate artifacts (file summaries, commit summaries, narrations) on disk to keep runs fast and cheap. Key Differentiators Reason-aware – Supports optional reasoning/thinking modes for both cloud (OpenAI o-series) and local (Qwen3, DeepSeek-R1) models. Multi-provider – Works with Gemini, OpenAI-compatible endpoints (Ollama, llama.cpp, LM Studio, vLLM, OpenRouter), and IBM Bob Shell. Deterministic + AI – Heuristic pre-filtering and deterministic post-processing surround the LLM calls, ensuring reproducibility and control. Local-first – Can run entirely offline with local models via Ollama or llama.cpp. Built with Tauri – A lightweight desktop app (Rust backend, web frontend) that works on macOS, Windows, and Linux. Target Users & Impact Git Lens is for engineering teams, open-source maintainers, and anyone who needs to rapidly understand an unfamiliar codebase. It reduces onboarding time from weeks to hours, surfaces institutional knowledge buried in commit history, and provides a reusable artifact that can be shared across the team.