Show HN: Agentic semantic search, but with GitHub APIs (github.com)

An open-source tool demonstrated on Hacker News implements an agentic semantic search that queries GitHub and GitLab via their APIs instead of crawling a local filesystem. Intended for product managers, business users, and developers who need quick, high-level understanding of unfamiliar codebases, it wraps API calls as "tools" for an LLM (default gpt-5) and prompts the model to use those network tools to fetch and reason over repository content. The project started as an experiment in storymachine but reportedly works well for routine code exploration, supporting both public and private repos (via GitHub/GitLab tokens) and MIT-licensed distribution. Technically, the system issues parallel API calls (network-bound tool calls) to speed file access, and exposes both a CLI (ask-github) and a Python library. Key runtime knobs include --max-iterations (default 20) to control reasoning/search depth and --max-workers (default 15) for parallel fetches; you set OPENAI_API_KEY and optional GITHUB_TOKEN/GITLAB_TOKEN for auth. It can analyze branches, commits, and refs, and lets you choose model and temperature. The author notes it’s not optimized for efficiency and lacks formal benchmarks, but the agentic design—LLM orchestrating API tools—makes it a pragmatic way to perform semantic code search across remote repos with minimal setup.