KNOWLEDGE ASSISTANTS

Answer questions from your own documents.

Documents are chunked and embedded as they arrive. A question finds the right passage, and an agent writes the answer with sources you can open.

STEP 1

Ingest and embed

Point the stack at your documents. Text is split into chunks and embedded as the files arrive.

PDFs and text files alongside table data
Runs against a local embedding model

osdbstore

Knowledge assistants: documents indexed in the catalog1400 x 1000 pxDelete this block and drop in an Image or Video widget

psql

-- dense + BM25, RRF-fused
SELECT title, chunk, rrf_score
FROM llm.search('How does ColdFront read Iceberg tables?', 5);

STEP 2

Retrieve by meaning

Dense vector search and keyword ranking are fused into one result list, in one SQL call.

Source passage shown with every hit

STEP 3

Answer with an agent

A LangGraph agent drafts the answer, checks it against the passages it found, and records every step.

Step-by-step trace stored in Postgres
Cost shown for each run

psql

-- run an agent and read its trace
SELECT agent.submit('answer',
  jsonb_build_object('question', 'How does ColdFront read Iceberg?'), NULL) AS run_id;

osdbstore

Knowledge assistants: an answer with its sources1400 x 1000 pxDelete this block and drop in an Image or Video widget

STEP 4

Keep data where it is

Use a local model when the data must not leave your network, or a hosted one when you need more capacity.

Ask your documents a question.

Start with the one-command stack and the sample documents.