Expand one message with its full part bodies (tool_call / tool_result / reasoning / file), plus conversational neighbors for context. Pass a message_id from pond_search or a transcript line; context_before / context_after size the neighbor window (default 3, like grep -B/-A). For the whole session use pond_get_session. Response format details: resource schema://pond.

Read a whole past session as a chronological transcript - the tool for analyzing, reviewing, or summarizing a session (user/assistant text plus one-line tool/file refs; tool bodies stay one pond_get_message away). Pass the id from pond_search or a subagent footer; a message_id also works - it resolves to its parent session with the page anchored at that message. Paging: limit (default 20), from="end" reads the most recent turns first (the session's final state; late conclusions supersede early ones), after_message_id / before_message_id continue from a page marker. The first page lists subagent sessions in a footer - pass a listed id back to open one. Not for bulk export - use pond copy --to <file>. Response format details: resource schema://pond.

Find relevant messages in past sessions - the entry point for recall: "have we worked on X", "what did we decide about Y", "find the session where...". mode="vector" (default) matches meaning; mode="fts" matches exact whole words (BM25). Scope with project / session_id / source_agent / from_date / to_date; keep the query semantic (concepts, not project names). Returns scored hits grouped by session, best session first; pass a hit's session_id to pond_get_session or its message_id to pond_get_message to read it. Searches conversational text only (tool calls/results and reasoning are excluded by design - a gap there is expected, not a failure) and excludes subagent sessions; reach both via pond_sql. Response format details: resource schema://pond.

Advanced escape hatch: run ONE read-only SQL statement (SELECT/WITH, DataFusion / PostgreSQL-compatible) over the sessions / messages / parts tables. NOT for finding or reading conversations - pond_search and pond_get_session / pond_get_message cover almost all recall. Reach for SQL only for: aggregation (counts, group-by, joins, time buckets), exact strings or identifiers in conversational text (contains_tokens / fts), tool-call analytics and tool bodies, subagent sessions, bulk export (format=parquet|ndjson). Read resource schema://pond-sql FIRST - exact columns, indexed predicates, JSON access rules, worked examples; do not guess column names or JSON paths. Inline text output is row-capped and long cells clip with a +N chars marker (full values via format=parquet|ndjson); queries are wall-clock-capped (raise via timeout_seconds).