claude-telepathy

One brain for all your Claudes. Every Claude Code session — running, ended, or compacted — becomes a mind your current session can read.

claude-mem gives Claude a diary. telepathy gives your Claudes one brain.

Status: building v1 — compaction rescue, continuity briefs, recall. Design: DESIGN.md.

Why

• Compaction destroys your working context ("50 first dates", "the dumb zone") — and the full transcript is still sitting on disk with no pointer back to it.
• Every new session is an amnesiac — you re-explain decisions your other sessions already made.
• The fix requires nothing new: Claude Code already writes everything to ~/.claude/projects/. Telepathy indexes it and wires it back in through official hooks — automatically, zero config.

Install

npx claude-telepathy install   # wires hooks + backfills your existing history (seconds)

Restart running sessions once; every session after that participates automatically.

telepathy list                 # your sessions, all projects
telepathy find "redis retry"   # full-text recall across everything you and Claude ever did
telepathy uninstall            # removes hooks cleanly; keeps the index

What v1 does

1. Compaction rescue — PreCompact snapshots your working state; after compaction the session gets it back: task, decisions, files in flight, next step.
2. Continuity briefs — every fresh session starts knowing what the last session here did.
3. Recall & search — telepathy find / /telepathy recall answer "how did we fix that last week?" from your own transcripts.

Principles: hook-enforced (never prose Claude can ignore) · retrieval-on-demand (no token bloat) · zero files in your repos · transcripts are the store (nothing to build up).

Honest limitations

Bash-mediated edits are recorded coarsely. Per-machine (no sync yet). Sessions deleted by Claude's 30-day cleanup are indexed only if telepathy was installed before.

Unofficial community tool — not affiliated with Anthropic. MIT.