Anthropic — Claude

Every part of the factory that reasons runs on a large language model. The factory uses Anthropic's Claude as that reasoning engine — the model that reads retrieved evidence and writes grounded, cited clinical reasoning. It's an external API connection, one of the Frontier Models the factory reaches over the network.

Illustrative. Claude reads retrieved evidence and returns a cited answer — an external API, an honest 503 when absent.

Where it plugs in

Capability	What Claude does there
Precision Intelligence Engine	RAG interpretation — turns annotated variants into a cited, plain-language clinical narrative, and coordinates the agents.
All 8 Intelligence Agents	Each is a retrieval-augmented service: Claude reasons over the retrieved evidence to produce the agent's cited answer — a dose, a diagnosis shortlist, a trial match, and so on.
Workflow Composer	Turns a natural-language request into a validated, governed pipeline of capabilities.

In short, Claude is the reasoning behind the entire agent layer — the "intelligence" in intelligence agent.

Honest limits

• Decision support, never diagnosis. Claude's output supports a qualified clinician; it never diagnoses or prescribes on its own.
• Grounded, not free-associating. It reasons over retrieved evidence (RAG) with citations, and returns an honest degraded response (HTTP 503) when its knowledge base or API key is absent — it does not fabricate clinical content.
• An external dependency. Claude is a proprietary API from Anthropic; the specific model is configurable (the factory runs a current model such as Claude Opus 4.8). Nothing on this layer is mock-served.