Crow Literature QA

Crow — Scientific Literature Q&A (FutureHouse Platform)

Crow is FutureHouse's fast, production literature search agent. It is the deployed version of PaperQA2. Behind the scenes it searches Semantic Scholar, Crossref, and other indexes, fetches full text, runs the PaperQA2 retrieval-augmented generation pipeline, and returns a grounded answer with in-text citations.

Use this skill when the user asks a focused factual question about the scientific literature ("What is known about X?", "What dose of Y was used in study Z?", "Summarize evidence on …"). For broader review-style questions hand off to Falcon; for chemistry-specific questions hand off to Phoenix.

Prerequisites

• pip install edison-client (or uv pip install edison-client)
• EDISON_API_KEY in env. Get one from https://platform.edisonscientific.com/profile → Account → API Tokens. Free tier exists; deeper queries consume credits.

Minimal usage

from edison_client import EdisonClient, JobNames

client = EdisonClient(api_key=os.environ["EDISON_API_KEY"])
resp = client.run_tasks_until_done({
    "name": JobNames.LITERATURE,                  # alias for Crow
    "query": "What dose of semaglutide is used for chronic weight management in adults?",
})
print(resp.formatted_answer)                      # answer with [Author Year] inline citations

resp is a PQATaskResponse with:

Recipes

Batch many questions in parallel

import asyncio
from edison_client import EdisonClient, JobNames

async def main():
    client = EdisonClient(api_key=os.environ["EDISON_API_KEY"])
    queries = [
        "What is the half-life of remdesivir?",
        "What efficacy did GLP-1 agonists show in NAFLD trials?",
        "Latest evidence on creatine for cognitive function in older adults?",
    ]
    tasks = [{"name": JobNames.LITERATURE, "query": q} for q in queries]
    results = await client.arun_tasks_until_done(tasks)
    for q, r in zip(queries, results):
        print(q, "→", r.has_successful_answer, "\n", r.answer[:400], "\n")

asyncio.run(main())

Follow-up question on a previous answer

first_id = client.create_task({
    "name": JobNames.LITERATURE,
    "query": "How many species of birds are there?",
})
follow_up = client.run_tasks_until_done({
    "name": JobNames.LITERATURE,
    "query": "Of those, how many are corvids?",
    "runtime_config": {"continued_job_id": first_id},
})

Fire-and-poll

task_id = client.create_task({"name": JobNames.LITERATURE, "query": "..."})
# … do other work …
status = client.get_task(task_id)   # status.status: 'queued' | 'running' | 'success' | 'failed'

Picking the right FutureHouse agent

Failure modes

• has_successful_answer == False: the agent looked but couldn't ground a confident answer. Show the user resp.answer (often "I don't know" with reasons) rather than fabricating.
• 401 / Unauthorized: EDISON_API_KEY missing or invalid.
• Insufficient credits: tell the user to top up at https://platform.edisonscientific.com/profile.
• Long queries can take 30 s – 3 min. Use arun_tasks_until_done for parallel batches.