Docdex
Structural code indexer with deterministic LLM context
Start new thread
bekir dağ

Docdex - Structural code indexer with deterministic LLM context

by•
Docdex is a local-first indexing daemon for code and documentation. It combines ranked full-text search (Tantivy), AST-based symbol indexing, and a directed dependency graph to answer impact questions like “what breaks if this changes?”. It runs as a single lightweight service, supports multiple repos, integrates with local models via Ollama, and helps reduce LLM token usage by providing structured, persistent context.

Add a comment

Replies

Best
bekir dağ
Maker
📌
Thanks for checking this out. I built Docdex because I was spending too many tokens re-explaining my own codebase to LLMs. The goal wasn’t “another AI tool,” but a local, deterministic layer that digests structure (docs + AST + dependency graph) so models can focus on reasoning instead of rediscovery. I’m especially interested in feedback on: Structural indexing vs embedding-heavy RAG Long-lived daemon model vs per-call tools Practical token reduction strategies Happy to go deep on architecture or tradeoffs.