Stop wasting tokens and re explaining your project between sessions.
https://github.com/raiyanyahya/recall
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
The cold-start problem is real — I lose so much context re-explaining my project to Claude Code every session. Like that you kept it fully local, no API key, nothing leaving the machine. That privacy-first choice is what makes it actually usable for client work where you can't send code anywhere.
Curious how you handle the condensing — is the summary generated locally too, or is it more of a structured log you feed back in? Either way, solving this for subscription users (not API) is a smart niche.
thanks! the summary is generated locally using libraries .. the summary does not use llms and the goal is to save token usage when you clear the cache coming from a long running session.
Keeping the summary local is a strong constraint.
I would still separate session condensation from durable product decisions. A generated summary can explain what happened and what comes next, but it should not silently decide which assumption became product truth.
Have you considered producing two outputs: a replaceable resume state for the next session, and a small list of decisions that require explicit confirmation before they become durable?