Stop wasting tokens and re-explaining your project every session. Recall gives Claude Code durable memory — entirely offline.
Claude Code starts every session cold. Recall keeps a local log of your sessions and condenses it into a resume-ready summary — entirely on your machine. No API key, no external model, nothing sent anywhere. It's built for people running Claude Code locally on a subscription: the only AI in the loop is Claude Code itself; the summarization is done by a classical Python summarizer.
Cool approach, local-first summarization is underrated. Curious about one design choice though. Why a classical summarizer instead of exposing the memory as an MCP server to Claude Code directly?
I've been building MCP servers lately and the pattern that works well for memory is: store sessions as structured chunks locally, expose them via FastMCP with tools like search_memory, get_session, recall_decisions. Claude Code calls them on demand instead of getting a prepended summary. Saves more tokens because you only pull what's relevant for the current task, not the whole resume.
The tradeoff is your approach has zero latency on session start and works without any MCP setup, which is genuinely simpler for people who just want it to work. So depends on the user. Heavy projects with 50+ sessions probably benefit from retrieval, smaller projects from your summary approach.
One thing I'd add fwiw, decisions and gotchas are way more valuable to carry across sessions than general progress. The "we tried X, didn't work because Y" lines are what saves real tokens. Are you tagging those separately or treating all session content equally in the summary?
local summarization because the goal is to save tokens :)