Skip to content
Agentic WorkFlow

Q&A Agent

Q&A Agent

Section titled “Q&A Agent”

Generate an action plan and execute it. Can use external tools.

What it does

Section titled “What it does”

Answers questions using a knowledge base by retrieving relevant context and generating responses.

When to use it

Section titled “When to use it”

Use this agent to build question-answering systems that reference specific documents or knowledge bases.

Run on-device

Section titled “Run on-device”

What you connect to the Model port decides the engine — there is no setting to switch:

  • A chat model (e.g. Chat OpenAI, Web LLM) — retrieves passages from the connected Knowledge and writes an answer from them.
  • A Local Q&A Model — answers in your browser: free, private, offline. It retrieves the top passages from the connected Knowledge and quotes the best-scoring answer span, so the answer always comes verbatim from your documents. System message and Output Parser don’t apply.

The on-device path needs a Knowledge: an extractive model can only quote from a passage, so connect a Local Knowledge (pair it with Local Embeddings to keep everything on-device).

Inputs and settings

Section titled “Inputs and settings”
SettingNotes
Options → System messageExtra instructions appended to the Q&A prompt (chat model only).
Options → Knowledge documents (k)How many passages to retrieve from the Knowledge (default 4).
Options → Max knowledge context lengthMaximum characters of retrieved knowledge to use (default 4000). On-device, each passage is capped to this length.
Options → Fail on knowledge errorChat model only: fail when retrieval errors instead of answering without context (default off).

Outputs

Section titled “Outputs”
  • Chat model — returns { response }, the generated answer.
  • Local Q&A Model — returns { response, answer, score, source }: response and answer are the quoted answer (empty if none was found), score is the model’s confidence, and source names the document it came from (when its metadata has a source, file name, title, or URL).

Dependencies and credentials

Section titled “Dependencies and credentials”
  • Model (required) — a chat model, or a Local Q&A Model to answer on-device.
  • Knowledge — optional with a chat model, required with a Local Q&A Model.
  • Output Parser (optional) — chat model only.

Example workflow

Section titled “Example workflow”

Connect Local Embeddings → Local Knowledge → the Q&A Agent’s Knowledge port, and a Local Q&A Model to its Model port. Index your documents into the knowledge store, then send questions to the agent — the answer and its source document come back without anything leaving the browser.

Troubleshooting

Section titled “Troubleshooting”
  • ”… it needs a context: connect a Knowledge to the Q&A Agent.” — the Local Q&A Model is connected but no Knowledge is.
  • Empty answer on-device — no retrieved passage contained an answer (a warning is logged). Raise Knowledge documents (k) or rephrase the question.
  • “Knowledge retrieval failed: …” — the knowledge store couldn’t be searched; check the Local Knowledge node and its embeddings.
  • If it calls an external service, verify credentials, permissions, and rate limits.
Section titled “Related nodes”