Answers - AiQL Handbook

Answer models handle natural-language queries over already ingested workspace data. They call workspace tools and reply with reasoning over that layer, not a generic model call on the raw file. They serve queries on the Middleware API and Explore in Observatory. Retrieval uses embeddings.

Parameters

Answer models are the same sparse mixture-of-experts family as pipeline: multimodal, with a vision encoder. Queries reason over workspace knowledge and can take images in the same turn.

Parameter	Value
Kind	Sparse mixture-of-experts, vision-language
Parameters	35 billion total, 3 billion active per token
Context	262k tokens native
Output budget	Up to 32k tokens per call
Input	Text and images
Tools	Search knowledge, inspect sources, and navigate related records

Configuration

Callers set these on a query. AiQL selects the checkpoint behind each mode.

Setting	How it is configured
Ask	Default mode. Direct answers over workspace knowledge.
Reason	Higher-effort reasoning. Used when the question needs a longer chain.
Build	Authoring mode. Used when the query should produce or extend a work product.
Visualize	Optional. Exposes graph visualization tools in the same turn.
Interactive	Optional. Exposes follow-up questions in the same turn.
Tools	The model may search knowledge, inspect sources, and navigate related records. It does not replace Studio review.