Return both a concise answer (when possible) and detailed supporting chunks.

IMPORTANT: The API key already selects the correct project/company.
Do NOT ask the user "which company" unless the user explicitly asks to compare multiple companies
or the KB contains multiple distinct entities and the question is genuinely ambiguous.

This is designed for fact-like queries (office address, phone, email, website) where the agent wants
a short response but still needs citations.
Prefer this for direct user questions; it is faster than listing documents.

Delete a document and remove its vectors from Qdrant.

Get a document (raw text + metadata) by id.

Prefer kb_search/kb_answer first. Use this only if you truly need full raw text.

Fetch a single URL, extract readable text, and ingest into the project KB determined by the API key.

List documents for the project determined by the API key.

Use this only for debugging/auditing (e.g., to see what sources exist).
For answering user questions, prefer kb_search / kb_answer.

This is intentionally paginated; increase limit/offset only when necessary.

Reindex a single document (re-chunk + re-embed + upsert vectors).

Search the project KB determined by the API key.

IMPORTANT: The API key already selects the correct project/company.
Do NOT ask the user "which company" unless the user explicitly asks to compare multiple companies
or the KB contains multiple distinct entities and the question is genuinely ambiguous.

Use this FIRST for almost all questions (services list, pricing, contact info, etc.).
It returns the most relevant chunks/snippets quickly.

Do NOT call kb_list_documents to answer questions; listing is for auditing/debugging.
Only fall back to kb_get_document when you truly need full raw text.

Upload raw text (or markdown) into the project KB determined by the API key.

Extract and save memories from a conversation for a given `user_id`.

Send recent chat `messages` (or a single text). The server will detect useful
long-term information (preferences, facts, important context) and store it as
searchable memories for that user within the project selected by `api_key`.

List user ids that have memories attributed to the given `agent_id`.

Return a summary of agent ids observed in memory metadata for the project.

Delete a memory by id (scoped to the project selected by the API key).

Delete a memory by id, enforcing project scoping via the API key.

List all memories for a specific `user_id` within the selected project.

List memories created by a specific agent.

Set `user_id` to narrow down to a single user.

Save a single memory item for the project selected by the API key.

Use this to store durable facts/preferences/events about a specific user.

Search stored memories for the project selected by the API key.

Returns the most relevant memory items for a natural language query.
Optionally filter by `kind`.

Search memories restricted to one or more users.

Use `user_id` (single) or `user_ids` (multiple) to constrain the search.
This is useful for agent workflows where you must not leak data across users.

Update the text (and optional metadata) for an existing memory.