Knowledge

Directory examples/python/features/knowledge

Try document ingestion, cited BM25 retrieval, listing and deletion through four capabilities. The node uses stored text without a model provider, so it runs against a local daemon offline.

Agent node label knowledge_desk, version 1.0.0.

Capabilities

Capability	Kind	Signature	Module	Summary
`ask`	traversal	`(question: str, top_k: int = 5, kind: Optional[str] = None) -> Answer`	`examples/python/features/knowledge/agent.py`	Ranked passages with citations and an extractive answer; ``kind`` filters on the
`catalog`	traversal	`() -> dict`	`examples/python/features/knowledge/agent.py`	Scope statistics and every stored document.
`forget`	traversal	`(doc_id: str) -> dict`	`examples/python/features/knowledge/agent.py`	Delete one document and its index rows.
`ingest`	traversal	`(documents: List[Dict[str, Any]]) -> dict`	`examples/python/features/knowledge/agent.py`	Write documents atomically. Each item: ``doc_id``, ``title``, ``metadata`` and

Commands

Taken from this example's own README and module docstrings.

python -m atfgraph.worker --module examples.python.features.knowledge.agent:app
python -m examples.python.features.knowledge.call

Files

examples/python/features/knowledge/README.md
examples/python/features/knowledge/__init__.py
examples/python/features/knowledge/agent.py
examples/python/features/knowledge/agentnode.yaml
examples/python/features/knowledge/call.py
examples/python/features/knowledge/fixtures.py

README

Rendered from examples/python/features/knowledge/README.md; the source file is not modified.

Try document ingestion, cited BM25 retrieval, listing and deletion through four capabilities. The node uses stored text without a model provider, so it runs against a local daemon offline.

Traversal	Surface	Notes
`ingest(documents)`	`Knowledge.put_many`	atomic batch; `text` is chunked server-side (1200 chars, 100 overlap), `chunks` keeps caller pages and per-chunk metadata
`ask(question, top_k, kind)`	`Knowledge.search`	BM25 hits with `doc_id#ordinal` citations; `kind` is an exact-match filter on document metadata; the answer is extractive with `[n]` markers
`catalog()`	`Knowledge.stats`, `Knowledge.documents`	scope statistics and a cursor-paginated listing
`forget(doc_id)`	`Knowledge.delete`	tombstones the document and rewrites the index in the same transaction

Documents use global scope by default. Inside a traversal, knowledge.scope("session") selects its session scope for per-session documents.

python -m atfgraph.worker --module examples.python.features.knowledge.agent:app
python -m examples.python.features.knowledge.call

The driver ingests invented travel-policy fixtures, asks three questions, lists documents, deletes one and shows the resulting ranking change. Scores measure lexical relevance; doc_id#ordinal citations identify chunks. Metadata filters limit which documents a search uses.