// the find
Storia-AI/sage
Chat with any codebase in under two minutes | Fully local or via third-party APIs
Sage is a RAG tool for querying codebases via chat, supporting both fully local (Ollama + Marqo) and cloud (OpenAI, Anthropic, Pinecone) backends. It indexes a repo into a vector store, then lets you ask questions about how the code works. Aimed at developers onboarding to unfamiliar codebases.
The dual local/cloud backend design is well thought out — you can run entirely air-gapped with Ollama and Marqo, which most similar tools don't bother with. The benchmarks directory with documented retrieval experiments is genuinely useful; they actually measured chunking strategies, embeddings, and rerankers rather than just shipping defaults. The codebase is modular with abstract base classes for embedders, vector stores, and LLMs, so swapping providers doesn't require rewriting business logic. CI exists and there are real tests covering chunking and vector store logic.
Last commit was November 2024 — this project appears stalled, which is a real concern if you depend on it. The 'two minutes' claim in the description is marketing; indexing a large repo against a remote vector store can take considerably longer. The local stack requires running Marqo separately (a full Docker container), so 'fully local' has more setup friction than the README implies. No streaming responses in the chat interface, which makes it feel sluggish on longer answers.