AI Agents

Two downloadable artifacts help you connect an LLM agent to the Huli Public API without hand-authoring boilerplate. Both are regenerated from the API contract on every docs build, so they stay accurate as the API evolves.

#System prompt

A model-agnostic system prompt that primes any LLM (Claude, GPT, Gemini, …) with what it needs to call the Huli Public API correctly: the SMART on FHIR R4 model, authentication, the system/<Resource>.<perms> scope grammar, the available FHIR resources and their interactions, the OperationOutcome error shape, and cursor pagination. Drop it into your agent's system message verbatim.

Download: /huli-system-prompt.md

#OpenAPI toolset

Tool / function definitions derived directly from the Huli Public API OpenAPI specification — one entry per API operation. The JSON object exposes two top-level keys so you can paste the matching block into your provider's request:

• openai — an array of { type: "function", function: { name, description, parameters } } objects for the OpenAI / Azure OpenAI tools parameter.
• anthropic — an array of { name, description, input_schema } objects for the Anthropic Messages API tools parameter.

Each tool's parameter schema lists the operation's path and query parameters (with types and which are required); operations with a request body expose a single body object property naming the FHIR resource. The schemas are intentionally shallow — they give the model the operation names and argument shapes, not a full FHIR expansion. Pair them with the API reference and the system prompt above for field-level detail.

Download: /huli-openapi-toolset.json

#See also

• /llms.txt — a pointer index of every documentation page, for retrieval-augmented agents.
• /llms-full.txt — the entire docs corpus concatenated into one file.