Starter Agents

Use the Amass public APIs as an agent — an LLM-driven loop that calls BiomedCore and TrialCore as tools in response to natural-language questions.

The Amass Platform Starter Agent is a complete, runnable Python reference implementation. It performs agentic retrieval out of the box and ships with a router loop, conversation history, and a fallback chain across LLM providers. Use it as a starting point for your own agents, or as a reference for wiring the Amass APIs into an existing agent harness.

Quickstart

The agent is built with BAML for type-safe LLM function calls and managed with uv.

# 1. Install dependencies
uv sync

# 2. Configure environment variables
cp .env.example .env
# fill in AMASS_API_KEY and at least one LLM provider key

# 3. Generate the typed BAML client (required before first run)
uv run baml-cli generate

# 4. Start the interactive agent
uv run python -m src.main

You can also seed the first turn directly:

uv run python -m src.main "Find recruiting Phase 3 lung cancer drug trials"

You will need an AMASS_API_KEY — see Authentication for how to create one.

Make it your own

A few ways to take the starter further:

1. Run as-is to explore what the public APIs can do from natural language.
2. Use your preferred LLM — works with Anthropic, OpenAI, Google AI, or any model reachable via LiteLLM. Pair a fast model for routing with a strong one for synthesis, and chain providers into a fallback sequence for production resilience. See the repo for setup.
3. Tune the agent loop — adjust the router prompts, tool selection, or end-of-turn synthesis to fit your domain.
4. Lift just the HTTP wrapper — src/amass.py is a self-contained async client for the Amass platform APIs. Drop it into an existing agent harness and define your own tools around it.

For the raw API contract that any agent's tools should follow, see the LLM Quick Reference.

Typescript

As an alternative to the Python starter agent, you can also use the Starter Agent in TypeScript.