Claude Code Forgets Your Project Every Session? This Plugin Stores Context Locally

You know that feeling: you open Claude Code to continue yesterday's work, and it has zero memory of your project. You're back to square one — re-explaining the architecture, the decisions you made, where you left off. It's like working with a new colleague every single session.

That's the problem Recall solves.

The approach is clever: instead of calling an external LLM to summarize your sessions (which would cost tokens and raise privacy concerns), it uses a classical Python algorithm (TextRank) to condense your conversation history into a compact context.md file. Next session, Claude Code reads that file first and remembers your project instantly.

The key selling point: everything stays on your machine. No API keys, no external model, no data sent anywhere. For privacy-conscious developers, this matters.

Technically, it runs as a Claude Code plugin. After installation, it silently saves context at the end of each session and loads it at the start of the next. You barely notice it's there — until one day you realize Claude Code actually remembers that architecture decision you discussed three days ago.

Why is it trending now? Claude Code's user base is growing fast, and this pain point hits every heavy user. Some folks have compared it to Cursor's built-in memory — the consensus seems to be that this local approach is more transparent and gives you more control.

The only requirement: you need Claude Code installed locally. But if you're already using it, this plugin is basically mandatory. The tokens and explanation time it saves you will pay back the 2-minute setup cost many times over.

git clone https://github.com/raiyanyahya/recall.git && cd recall && pip install -e .