GitHits beta 0.9
Give your AI coding agent access to open-source code
Start new thread

GitHits beta 0.9 - Give your AI coding agent access to open-source code

by•
GitHits gives coding agents access to the open-source code your app depends on. Get real implementation examples, dependency source navigation, package inspection and documentation. Agents can grep and read your codebase. They can't grep and read the open-source code your app depends on. That's where they start guessing, retrying, and looping. GitHits builds a version-aware index on demand. Agents can search, navigate, and inspect the code behind their dependencies. CLI: npx githits@latest init

Add a comment

Replies

Best

The dependency-context angle makes a lot of sense. Docs explain the API, but source code and real examples show how it actually behaves. I’d be interested to see how GitHits weighs tests vs production implementations when finding examples.

Maker

 We don't automatically favor production code over tests.

Our ranking includes signals for both because tests often demonstrate the intended usage pattern much more clearly than production implementations. Production code shows how things are used in the wild, while tests often isolate the behavior you're trying to understand.

Both can be valuable depending on the query.

Giving agents real OSS context feels like a cheat code. Curious how you rank which repos are signal vs random GitHub soup.

Maker

 For the distilled examples, ranking is one of the hardest problems.

We don't treat every repository equally. We combine code relevance, semantic relevance, repository quality signals, and package metadata to separate useful implementation patterns from random GitHub noise.

The goal is to surface examples that show how similar problems are actually solved in code, rather than repositories that simply contain matching keywords. However, sometimes the solution to an issue can be found in a single random repo, which is like finding a needle in a haystack.

giving the agent access to open source code is the easy half. the hard half is helping it pick the right code. github has incredible code and also a long tail of broken half written experiments. how do you weight signals like maintained recently, used by many, tests pass, vs raw similarity to what the agent is trying to write?

Maker

Here's a press story worth checking out:

Selected quotes:

"Analysis of 16 widely used code-generating LLMs across 576,000 generated code samples found that 21.7% of the package names referenced in AI coding outputs were hallucinations — meaning no such packages existed in npm or PyPI. That’s nearly one in four dependencies simply invented by the model. GitHits is betting that plugging this specific gap — open-source code context for AI agents — is the infrastructure play that the entire coding-agent ecosystem has been missing."

"The origin story is refreshingly unglamorous."

"The GitHits Google for code framing isn’t just clever marketing — it’s a deliberate competitive positioning. As Heinisuo puts it: 'OpenAI, Anthropic, and Google have left a gap in the market. GitHits doesn’t compete with Codex, Claude Code, or Cursor, but complements them by bringing open-source code as context for agents to end retry loops and reduce token consumption.'"

"GitHits raised €1.5 million in a pre-seed round led by Vendep Capital, with participation from Trind VC and angel investors including Peter Sarlin, Zach Shelby, and Jerry Liu. That last name is notable — Jerry Liu is the co-founder of LlamaIndex, one of the most widely used frameworks for building LLM-powered applications. His backing signals more than capital; it signals conviction from someone who lives inside the AI agent stack daily."

First
Previous
12