Show HN: Give (low vRAM) local sLLMs codebase-wide context with SQLite (attempt)

Category: ai-ml

Tags: local-llm, code-indexing, tree-sitter, sqlite, ollama, ai-coding-agent

Score: 5.8/10 (Innovation: 6, Technical: 6, Documentation: 6, Utility: 5)

arcKode is a local-first AI coding agent that uses Tree-sitter to build a symbol graph of your codebase, stored in SQLite with FTS5, enabling codebase-wide context for small local LLMs via Ollama. It combines incremental indexing, token-aware context packing with reciprocal-rank fusion, and anchored patch application with snapshots. Interesting for demonstrating how structured code indexing can meaningfully stretch limited VRAM budgets for local agentic coding.

Target audience: backend devs, developers running local LLMs, privacy-conscious engineers

Repository: https://github.com/govindgcoder/arcKode · Python · AGPL-3.0 · 2 stars

View on Hacker News