Intelligence Core
The Agentic Orchestrator's model roles and what the reasoning model sees on each call.
Controlled Availability
The Agentic Orchestrator is in Controlled Availability and comes to your Enterprise Assistant in Q1 2027. Capabilities, limits, and interface labels described here may change before general availability. To request access, contact your Moveworks account team.
One reasoning model runs the Orchestrator’s loop: it plans, calls tools, writes search queries, reads images, and writes every reply. A small safety classifier and a lighter web search model work beside it.
How It Works
- Reasoning model. Each call carries the base prompt, your instructions, the Skill catalog, the user’s profile, the user’s local date and time, and the callable tools, plus plan state and session variable names; see Memory. If the primary model is unavailable, the call falls back to ordered alternates.
- Reasoning effort. Starts at low. The model raises it to medium or high before a complex plan or a hard failure, then lowers it when that step is done.
- Safety classifier. A small model checks each user message while the turn runs. Nothing reaches the user until the check passes, and a blocked message gets a short fixed reply instead.
- Search. Web search is a separate call to a lighter model, using the user’s approximate location; it returns titles, links, and snippets. Enterprise search adds no model: the reasoning model writes the query plus up to two alternative phrasings and reads the ranked results directly.
- Images. The reasoning model is multimodal. For a request about an attached image, it loads the image and reads it directly. Enabled per organization.
Limits
Configure
To steer when the assistant searches the web or looks at an image, say so in the instructions, for example: “If the user attaches a screenshot of an error, look at it before answering.”