Quickstart

pond init  # interactive
pond init -y --every 5m  # non-interactive: enable every discovered adapter, sync every 5m

init walks storage, adapters, MCP registration, and an optional schedule, then writes config.toml in one pass. An interactive run offers to run the first sync in the foreground with full progress, and registers the schedule once that sync completes. Idempotent - re-run to repair. -y takes defaults; --every is opt-in (-y alone never schedules).

pond sync  # import and index - every enabled adapter

sync only ingests already-enabled adapters; it never enables on its own. Manage the set with pond adapters list|discover|enable|disable.

Discovery looks under your home directory on every OS - ~/.claude/projects, ~/.codex/sessions, and the rest live in the same place on Windows (%USERPROFILE%\.claude\projects, ...); only the Claude Desktop app's sessions sit elsewhere (%APPDATA%\Claude on Windows). Nothing to configure either way - pond init and pond adapters discover look in the right places for the OS they run on.

Then confirm it worked:

pond status  # stored sessions/messages, per-adapter pending counts, index readiness

It searches your history through pond:

• "find how we fixed the OCC retry race"
• "what did we decide about the storage substrate, and why?"
• "pick up where we left off on the tokenizer experiment"

Search spans every client - including sessions made in a different tool. Re-run pond sync (or let the schedule) to pick up new sessions.