# How I built local-first memory for Claude Code, Cursor, and Codex - 94.5% LoCoMo recall@10, 70ms p50

Page: https://stenobird.com/podcast/programming-tech-brief-by-hackernoon-6364125/how-i-built-local-first-memory-for-claude-code-cursor-and-codex-94-5-locomo-recall-10-70ms-p50
Text version: https://stenobird.com/podcast/programming-tech-brief-by-hackernoon-6364125/how-i-built-local-first-memory-for-claude-code-cursor-and-codex-94-5-locomo-recall-10-70ms-p50.md
Podcast: [Programming Tech Brief By HackerNoon](https://stenobird.com/podcast/programming-tech-brief-by-hackernoon-6364125)
Published: 2026-06-02T16:00:53+00:00
Episode link: https://share.transistor.fm/s/e7f1303f
Audio file: https://media.transistor.fm/e7f1303f/e97f4bc1.mp3
Processing state: not_requested
JSON: https://stenobird.com/v1/public/podcasts/programming-tech-brief-by-hackernoon-6364125/episodes/how-i-built-local-first-memory-for-claude-code-cursor-and-codex-94-5-locomo-recall-10-70ms-p50
Duration seconds: 1120

## Resource

This story was originally published on HackerNoon at: https://hackernoon.com/how-i-built-local-first-memory-for-claude-code-cursor-and-codex-945percent-locomo-recall10-70ms-p50 . Open-source local-first memory for Claude Code, Cursor, Codex via MCP. 94.5% LoCoMo recall, 70ms p50, no API keys. Five techniques explained with code. Check more stories related to programming at: https://hackernoon.com/c/programming . You can also check exclusive content about #open-source , #ai , #programming , #machine-learning , #python , #large-language-models , #mcp , #developer-tools , and more. This story was written by: @oleksiijko . Learn more about this writer by checking @oleksiijko's about page, and for more stories, please visit hackernoon.com . PMB is an open-source MCP memory server for Claude Code, Cursor, and Codex. It runs 100% locally (SQLite + LanceDB), needs zero API keys, and hits 94.5% LoCoMo recall@10 at 70ms p50 - matching or beating cloud alternatives like mem0, Letta, and Zep. This post walks through the five techniques behind it: predicate-aware reranking, verb synonym expansion, no-LLM atomic fact extraction, pattern query splitting, and a durable async embed queue - with the actual code from the repo.

## Actions

- request_transcript: `POST https://stenobird.com/v1/public/podcasts/programming-tech-brief-by-hackernoon-6364125/episodes/how-i-built-local-first-memory-for-claude-code-cursor-and-codex-94-5-locomo-recall-10-70ms-p50/transcription-requests` — Idempotently request low-priority transcript generation for this episode.
- read_markdown: `GET https://stenobird.com/podcast/programming-tech-brief-by-hackernoon-6364125/how-i-built-local-first-memory-for-claude-code-cursor-and-codex-94-5-locomo-recall-10-70ms-p50.md` — Read the agent-friendly Markdown representation of this episode resource.

A page view does not enqueue transcription. Agents should invoke `request_transcript` explicitly when they need this episode processed.

## Transcript

Full transcripts are not published on public pages unless there is a clear rights basis.