Show HN: Memory Lane – bootstrap your naive Claude instances with their history (github.com)

Memory Lane is an open-source persistent memory system designed to “bootstrap” naive Claude LLM instances by capturing, archiving and exposing conversation histories so agents can maintain continuity across context windows and sessions. Built for the magent agent used by cryptograss, it ingests Claude JSONL logs (extendable to other formats), stores conversations in PostgreSQL, and serves memory to models via an MCP (Model Context Protocol) server. A Django web app provides a memory viewer, while a watcher process imports logs in real time; importers/parsers and a scrubber handle format parsing and secret sanitization. The project is MIT-licensed. Technically, Memory Lane models memory as eras (major phases), context heaps (groups of messages within a window), individual messages (turns, tool uses, results), and compacting actions (when contexts are summarized). The MCP server exposes tools like bootstrap_memory (load recent context, era summaries, reflections), search_messages, get_recent_work, random_messages, and get_era_summary, enabling retrieval-augmented continuity and contextual bootstrapping for Claude Code. Setup is standard Django/Postgres (venv, migrations, runserver, watcher/docker-compose); CLI utilities support JSONL import, chain repair, analysis and backups. For AI/ML practitioners this offers a practical, extendable way to give stateless LLMs durable, queryable memory and better multi-session behavior while retaining controls for privacy and data hygiene.