Local-first search across every AI coding session on your machine.

Jump between Claude Code, Codex, and whatever comes next; Recall pulls those scattered local sessions into one searchable index, tracks usage when token metadata is available, and drops you back into the original CLI.

brew install samzong/tap/recall
# or
make install # from a source checkout

Generate a static completion script and load it from your shell fpath:

recall completions zsh > ~/.zsh/completions/_recall

Add to ~/.zshrc (works alongside gmc and other tools):

fpath=(~/.zsh/completions ~/.grok/completions/zsh $fpath)
autoload -Uz compinit && compinit

Other shells:

recall completions bash > ~/.bash_completion.d/recall
recall completions fish > ~/.config/fish/completions/recall.fish

One index across every AI coding CLI. Sync once, search everywhere, resume right where you left off.

recall sync          # incremental sync (safe to run anytime)
recall sync --force  # reprocess every session (after changing embedding model)
recall               # launch TUI
recall search Q      # one-shot CLI search
recall search Q --project /path/to/repo
recall usage         # usage dashboard
recall usage --json  # usage report for scripts
recall export --source codex --project /path/to/repo --limit 20 > recall-export.jsonl
recall import recall-export.jsonl --dry-run  # preview an import
recall session list --source codex --limit 20  # list sessions for agents/scripts
recall session show --id <session-id> --include metadata,messages --format json
recall session export --id <session-id> --format jsonl
recall session share --id <session-id> --format json  # publish one selected session
recall session resume --id <session-id> --print-command
recall info          # index stats and worker status

recall export writes JSON Lines to stdout, with one JSON object per indexed session. Redirect stdout to save an export file:

recall export --source codex --project /path/to/repo > recall-export.jsonl

Each session includes schema_version, record_type, session, messages, usage_events, and events. It covers the portable session data Recall can import again; derived index state such as FTS rows, embeddings, and background job state is rebuilt locally. Optional fields are emitted as null. By default all sessions are exported; use --limit N to truncate. --time filters on started_at, so prefer a full export when moving data between machines.

recall import <file> (or - for stdin) loads sessions from an export file into the local index. How the file travels between machines is up to you.

# machine A
recall export > recall-a.jsonl
# machine B
recall import recall-a.jsonl

• Idempotent: a session whose (source, source_id) already exists locally is skipped, so re-running an import is safe and local data always wins.
• Imported sessions are searchable and appear in usage reports, but cannot be resumed on this machine (the source tool's own files were not copied); the TUI explains this if you try.
• --dry-run parses and reports counts without writing anything.

MIT

Thanks to tokscale for the usage dashboard reference and token accounting behavior.