Search the local chat archive and return matching transcript chunks. Use this first to recall past work on a topic; then read a hit with get_chat_logs. For extracted action items, comparisons or briefs use compile_project_insights instead. Methods: 'semantic' (default) ranks chunks with Okapi BM25 plus recency decay and returns [{file, score}]; 'keyword' returns chunk ids containing every keyword; 'date_range' returns [{file_name, last_modified}] for chats saved between start_date and end_date; 'related' returns [{file, score}] for chats similar to file_name. Chunk ids look like 'name__part_2.json' and can be passed to get_chat_logs as chat_id. Active handoff receipts are ranked first. Read-only for chat data; it refreshes the local search index cache. An empty archive returns a single placeholder entry such as 'No stored chats found.'.

List archived chats, or read one chat as a message range, a short summary, or metadata. Use it after search_history to open a hit, or with no chat_id to browse the archive page by page. To search by topic use search_history instead. Without chat_id: returns a page header '[Page N of M (Total chats: T)]' followed by file names. With chat_id: view_type 'content' returns messages start_msg..end_msg as plain text with a '... of N total' header, truncating each message to max_msg_len characters and collapsing code blocks to one-line summaries unless summarize_code=false; 'summary' returns the opener and closer; 'metadata' returns {file_name, title, message_count, file_size_bytes, last_modified, saved_at, client}. Older chats are shown pruned to save tokens. Keep ranges small (the default is 20 messages): max_msg_len=0 with a wide range can return a whole transcript. Read-only.

Import conversations into the archive from pasted content, a local file, or an editor's transcript folder, normalizing ChatGPT, Claude, Cline/Continue JSON, JSONL and Markdown formats. Use it to bring outside history in; to save the current conversation use manage_session_state(action='save'). source_type 'raw_content': payload is the chat JSON (string or object); 'local_path': payload is a file path inside the archive, the current directory, the temp directory or a configured transcripts folder (other paths are refused); 'agent_transcripts': scans the transcripts folder for 'client' ('cursor', 'continue', 'cline', 'copilot', or source_dir) and imports up to 'limit' newest files, skipping ones already imported; 'cursor_agent_transcripts' is a deprecated alias for Cursor. Writes new JSON files into the archive and updates the search index; never deletes. Imports with an existing title overwrite that chat. Returns a confirmation string for single imports, or {imported: [{file, status, dest}], errors, total_scanned} for folder syncs.

Derive structured insight from archived chats: action items, a topic index, a two-chat comparison, or a Markdown project brief. Use it when you need a digest rather than raw text; to find chats use search_history and to read them use get_chat_logs. insight_type 'action_items': TODOs, checkboxes and 'next' tasks from file_name (or the first of target_chats), returned as a list of strings. 'knowledge_index': chats grouped by topic, returned as {topic: [files], built_at}, or file counts when summary_only=true (use that on large archives); rebuild=true rescans every file. 'compare_chats': shared and unique topics of file_name_a and file_name_b (or the first two target_chats) as Markdown. 'project_brief': merges summaries and action items of target_chats (or file_name) into exports/<brief_title>.md and returns its path. Side effects: 'project_brief' writes or overwrites an export file; 'knowledge_index' rewrites the cached index file. Chat files are never modified.

Housekeeping for the archive: compress old chats, find or remove duplicates, change settings, or report server capabilities. Use it for storage and configuration only, not for reading or searching chats. op_type 'compress': gzips chats older than days_old (default from settings, 30) in place and returns {compressed, skipped_recent, days_threshold}; compressed chats stay readable and searchable. 'deduplicate': finds chats with identical content; with dry_run=true (default) it only lists them, with dry_run=false it permanently deletes the duplicates, returning {dry_run, duplicates, unique_count}. 'configure': updates settings (see 'settings') and returns which keys changed. 'capabilities': read-only summary of tools, prompts, version and client paths. Destructive only for deduplicate with dry_run=false; ask the user before doing that.

Save, auto-save, merge, export or delete archived conversations, and watch the archive for changes. Use it to persist the current conversation; to import outside history use sync_workspace_data. action 'register_auto_save': remember conversation_name and messages so they are written when the client disconnects (call it early and re-call it to refresh). 'trigger_auto_save': write the registered session now. 'save': write messages to <conversation_name>.json, overwriting a chat with that name; below the auto_save_message_threshold (default 20) it is skipped unless force_save=true. 'merge': append new_messages to file_name, skipping duplicates. 'export_markdown': write exports/.md. 'delete': permanently delete file_name; refused unless confirm=true, so ask the user first. 'watch_folder': report {new_files, modified_files, total_tracked} since the last call (only updates the watch checkpoint). Returns a confirmation string, or the report object for 'watch_folder'.

Record what this agent run did and what the next agent must do, as a handoff receipt for the current workspace. Use it before stopping or handing work to another agent or model; read the current state with get_handoff_status. A receipt with status 'open' (the default) marks the workspace as blocked until a later receipt supersedes it with status 'closed' and a closure (closed_by, closed_at, closure_basis). It becomes stale automatically when any file listed in invalidation.stale_if changes after it was saved. Receipts are stored as new JSON files in the archive (never overwritten), indexed, and ranked first in search_history. Returns the saved file name and its receipt_id (pass that as 'supersedes' later).

Check whether this workspace has unfinished work handed off by an earlier agent. Call it at the start of a session on a project, before editing; to record a handoff use save_handoff_receipt. Returns {blocked, blocked_by_receipt, active_chain (newest first), newest_receipt_id, next_safe_action}; with no receipts it returns {blocked: false, reason, active_chain: []}. When blocked is true, follow next_safe_action (stale receipts tell you to re-verify first). Read-only.