Archiveye

Execute read-only SQL SELECT on one structured document in a collection. Use FROM t only. Requires prior get_tabular_schema. Arguments: - collection_id (int, required) - document_id (int, required) - sql (str, required): SELECT ... FROM t ... - sheet (str, optional): sheet slug or name for multi-sheet workbooks - limit (int, optional): max rows (default 1000)

execute_tabular_query

Full-text (BM25) keyword search inside ONE collection — best for exact terms, codes, names, or acronyms. Arguments: - collection_id (int, required): from `list_collections`. - query (str, required): keywords or short phrase. - limit (int, optional, default 5): max hits. Example: search_fulltext(collection_id=12, query="art. 32 GDPR", limit=5)

search_fulltext

Fetch metadata for a document that belongs to any of the user's collections (no body text). Arguments: - document_id (int, required): id of the document as returned by `search_semantic`, `search_fulltext` or `list_collections`. Returns a JSON object with id, metadata_id, name, type, size, processing_status, created_at, content_chunks and (when present) a `metadata` object from the metadata table (type, meta_description, json_data). When the document is structured tabular data, a `tabular` object includes the full sheets catalog (all columns per sheet) for multi-sheet workbooks — same data as get_tabular_schema. Use search tools for narrative text; use tabular tools for SQL analytics.

get_document

Return schema and stats for a structured document in a collection. Use when structured_document_count > 0 in list_collections. Arguments: - collection_id (int, required): from list_collections. - document_id (int, required): structured document id. - sheet (str, optional): sheet slug or name for multi-sheet workbooks. Call before execute_tabular_query.

get_tabular_schema

List all active knowledge base collections owned by this user, as a JSON array. Each entry contains: id, name, description, options (list of {name, value} pairs describing the collection's topics), document_count, indexing_status and last_indexed_at. Use this tool first and then pick the single most relevant collection_id to pass to `search_semantic` / `search_fulltext`. Example output: [ {"id": 12, "name": "HR policies", "description": "Internal HR rules, leave, onboarding", "options": [{"name": "topic_area", "value": "hr"}], "document_count": 34, "indexing_status": "COMPLETED", "last_indexed_at": "2026-04-15T09:12:03"} ]

list_collections

Retrieve the full markdown body of ONE document (exceptional use only). **Token cost:** returns the entire document text — use sparingly to avoid excessive token consumption. **Do not call routinely or in bulk.** Default to `search_semantic` / `search_fulltext` chunk previews; call this tool only when those previews are clearly insufficient to answer. Ask the user before using this tool unless they explicitly requested the full document text. Arguments: - document_id (int, required): from search results or prior tool output. Example: get_document_content(document_id=874)

get_document_content

Semantic (vector) search inside ONE collection — best for meaning, concepts, or paraphrased questions. Arguments: - collection_id (int, required): from `list_collections`. - query (str, required): natural-language query. - limit (int, optional, default 5): max hits. Example: search_semantic(collection_id=12, query="congedo parentale", limit=5)

search_semantic