Last updated: 2026-06-22
Short version: Recall collects nothing, sends nothing, and runs entirely on your machine. There is no server, no account, no telemetry, and no network connection of any kind.
Recall is a fully-local memory plugin for Claude Code. It reads your local
session transcript and writes its output into the .recall/ directory inside
your project. The two files it produces for you are:
.recall/history.md— an append-only log of each session (your prompts, Claude's replies, files touched, and commands run)..recall/context.md— a condensed summary of where you left off, produced by a classical Python summarizer that runs locally.
It also keeps small internal bookkeeping files in the same directory (for example, capture progress and a pause marker). Recall never writes outside this directory, and it refuses paths that would escape your project.
Recall processes the contents of your Claude Code session — prompts, replies, file paths, and commands — solely to write the two files above. This data:
- Stays on your machine. It is written only to the
.recall/directory in your project (the location is configurable viaoutput_dir). - Is owned and controlled by you. You can read, edit, or delete these files
at any time. Deleting
.recall/removes everything Recall has stored.
- No network access. Recall makes no HTTP requests and uses no networking libraries. It works fully offline.
- No telemetry or analytics. Nothing about your usage is collected or reported.
- No third-party services. Your data is never sent to any API, model endpoint, or external party — including any AI model. The only AI in the loop is Claude Code itself; Recall's summarization is a local algorithm.
- No accounts or keys. Recall requires no sign-up, API key, or credentials.
The only external program Recall invokes is your local git, read-only, to
record branch and diff information. It is configured so it cannot prompt for
credentials or contact a remote (GIT_TERMINAL_PROMPT=0), and you can disable
it entirely by setting include_git to false.
Because session transcripts can contain secrets, Recall applies best-effort
redaction (enabled by default via the redact setting) before writing its
files. It removes common patterns such as API keys, AWS access keys,
GitHub/Slack tokens, JWTs, private keys, and SECRET/TOKEN/PASSWORD/
API_KEY environment-variable values.
This is a safety net, not a guarantee. Treat the .recall/ files as you
would your source code. They are git-ignored by default; keep them ignored
unless you intend to commit them as shared team memory, and review them before
sharing.
Any changes will be reflected in this file in the repository, with the "Last updated" date above.
Questions or concerns? Open an issue at https://github.com/raiyanyahya/recall/issues