Save Claude tokens with semantic search powered by SQLite and Ollama

Just to clarify the background a bit. This project wasn’t planned as a big standalone release at first. On January 16, Ollama added support for an Anthropic-compatible API, and I was curious how far this could be pushed in practice. I decided to try plugging local Ollama models directly into a Claude Code-style workflow and see if it would actually work end to end. Here is the release note from Ollama that made this possible: https://ollama.com/blog/claude Technically, what I do is pretty straightforward: - Detect which local models are available in Ollama. - When…

Claude Code getting slower as your codebase grows? Ory Lumen is a local semantic search MCP server that cuts costs by up to 39% and sessions by up to 53%.

Hey HN! I built Retain as the evolution of claude-reflect (github.com/BayramAnnakov/claude-reflect). The original problem: I use Claude Code/Codex daily for coding, plus claude.ai and ChatGPT occasionally. Every conversation contains decisions, corrections, and patterns I forget existed weeks later. I kept re-explaining the same preferences. claude-reflect was a CLI tool that extracted learnings from Claude Code sessions. Retain takes this further with a native macOS app that: - Aggregates conversations from Claude Code, claude.ai, ChatGPT, and Codex CLI - Instant full-text…

Semantic search over your own Claude Code session history

Built this after realizing I was spending ~$1400/week on Claude Code with almost no visibility into what was actually consuming tokens. Tools like ccusage give a cost breakdown per model and per day, but I wanted to understand usage at the task level. CodeBurn reads the JSONL session transcripts that Claude Code stores locally (~/.claude/projects/) and classifies each turn into 13 categories based on tool usage patterns (no LLM calls involved). One surprising result: about 56% of my spend was on conversation turns with no tool usage. Actual coding (edits/writes) was…

As I started to use Claude Code to do more random tasks I realized I could basically build any CLI tool and it would use it. So I built one that controls the browser and open-sourced it. It should work with Codex or any other CLI-based agent! I have a long term idea where the models are all local and then the tool is privacy preserving because it's easy to remove PII from text, but I'd definitely not recommend using this for anything important just yet. You'll need a Gemini key until I (or someone else) figure out how to distill a local version out of that part of the pipeline. Github link:…