Recall Brings Local Project Memory to Claude Code Without an External Model
Recall is a new open-source Claude Code plugin that records project sessions locally and condenses them into a resume-ready context file using a classical Python summarizer instead of an external LLM call.
Recall is an open-source Claude Code plugin for local project memory. Its repository says it captures session history inside the project and writes a compact context.md summary with a local Python summarizer, so the memory workflow does not require an external model or API key. The project gained early attention through a Show HN post, making it a useful watchlist item for developers who restart AI coding sessions often.
Key takeaways
- Recall targets Claude Code users who repeatedly re-explain project state across sessions.
- The repository describes a local memory flow using
.recall/history.mdand.recall/context.md. - The summarizer is described as classical TF-IDF plus TextRank, with no LLM call for the summary step.
- The project states that transcripts stay on the user's machine, but users still need to review stored files before committing or sharing them.
- Hacker News and Trendshift both show early community interest, but this is not a substitute for a security review.
Practical LinkLoot angle
Recall is worth tracking because it addresses a practical failure mode in AI coding: losing working context after a session ends. Instead of paying tokens to rebuild state, the tool creates a local digest that can be loaded into the next Claude Code session. That is useful for solo builders, but teams need stricter rules around what gets logged, redacted, committed, and trusted.
| Tool or workflow | Best use | Limitation | Source |
|---|---|---|---|
| Recall | Local Claude Code project memory across sessions | Requires trust in local hooks and stored project files | GitHub repository |
CLAUDE.md | Stable project instructions and team conventions | Manual upkeep; does not automatically summarize prior work | Recall README comparison |
| Session resume or transcript replay | Full-fidelity continuation | Can be token-heavy and tied to local session history | Recall README comparison |
The sharp use case is long-running coding work with frequent restarts: migrations, refactors, bug hunts, and multi-day feature builds. A compact local summary can reduce repeated setup prompts, but the memory file should be treated as untrusted input when shared. If another contributor can edit context.md, they can influence the next session.
What to verify before you act
Review the repository before installing or enabling any hooks. Check what files are written, whether .recall/ is ignored or committed in your project, how redaction works, and whether the plugin can run commands during Claude Code lifecycle events. The README describes best-effort secret stripping, which is useful but not a guarantee.
Also test the quality of the generated summary on a non-sensitive project first. A local extractive summarizer can miss intent, stale decisions, or unresolved blockers. For production codebases, treat Recall as a context aid, not a source of truth, and keep final decisions in issues, pull requests, or documentation.
Source check
The GitHub repository confirms the core claims: local session logging, .recall/history.md, .recall/context.md, TF-IDF plus TextRank summarization, no external model call for the summary step, and best-effort redaction caveats. Hacker News confirms the Show HN launch discussion and early community attention. Trendshift provides an additional repository-momentum page, useful for visibility context rather than technical verification.
Recall is an open-source plugin that keeps local project memory for Claude Code sessions by logging history and generating a compact context file.
For more ways to compare agent memory, local tools, and workflow automation, see LinkLoot's guide to AI workflow automation.