Biomni – Open Responses SDK

For the open-source Biomni release (Apache 2.0).
This document shows how to call the OpenAI Python SDK Responses API for Biomni.

Prerequisites

  1. Sign up for a 20minds account
  2. Create an API key in settings
  3. Install Python 3.10+ and OpenAI Python SDK (tested with 2.15.0)
Install the SDK:
pip install openai==2.15.0

Configure the client

Biomni uses an API key passed via the OpenAI SDK api_key field.
from openai import OpenAI

TWENTYMINDS_BASE_URL = "https://api.20minds.ai/v1/biomni"
TWENTYMINDS_API_KEY = "your-api-key"

client = OpenAI(
    base_url=TWENTYMINDS_BASE_URL,
    api_key=TWENTYMINDS_API_KEY,
)

Create a response

model is optional and passed through to the Biomni A1 session (gpt-5-mini by default).
response = client.responses.create(
    input=[
        {
            "role": "user",
            "content": [
                {
                    "type": "input_text",
                    "text": "Suggest a biomarker panel for early-stage pancreatic cancer and explain why."
                }
            ],
        }
    ],
    model="gpt-5-mini",
)

print(response.id)
print(response.status)
print(response.metadata.get("biomni_session_id"))

Poll for completion

import time

resp = client.responses.retrieve(response_id=response.id)
while resp.status == "running":
    time.sleep(5)
    resp = client.responses.retrieve(response_id=response.id)

print(resp.status)
print(resp.output)
print(resp.usage)

Stream events

Biomni supports stream=True and emits OpenAI-style SSE events (for example response.output_text.delta, response.code_interpreter.*, and response.completed).
with client.responses.stream(
    input=[{
        "role": "user",
        "content": [{"type": "input_text", "text": "Run a quick analysis and show intermediate reasoning outputs."}],
    }],
) as stream:
    for event in stream:
        print(event.type)

Continue a conversation

Use previous_response_id to continue the same Biomni session.
followup = client.responses.create(
    input=[
        {
            "role": "user",
            "content": [
                {"type": "input_text", "text": "Now summarize that in 5 bullet points."}
            ],
        }
    ],
    previous_response_id=response.id,
)

Delete a response

client.responses.delete(response_id=response.id)
Deleting a response also deletes its underlying Biomni session.

Response shape (summary)

A completed response contains conversation items in output:
for item in resp.output:
    print(item["type"])
Notes:
  • output includes assistant message items with output_text.
  • If Biomni returns <solution>...</solution>, the server adds a function_call item named final_answer.
  • usage is normalized to OpenAI-style token fields and includes usage_usd.

File uploads

Biomni exposes OpenAI-style file endpoints under the same base URL.
from openai import BaseModel
from typing import Literal
import requests

class FileCreate(BaseModel):
    id: str
    response_id: str
    object: Literal["file"]
    filename: str
    status: str
    upload_url: str
    upload_expires_at: str
    created_at: str

file_record = client.post(
    "/files",
    body={"filename": "paper.pdf"},
    cast_to=FileCreate,
)

upload_response = requests.put(
    file_record.upload_url,
    data=requests.get("https://example.com/paper.pdf").content,
)
upload_response.raise_for_status()

response = client.responses.create(
    input=[{
        "role": "user",
        "content": [
            {"type": "input_text", "text": "Summarize the uploaded report."},
            {"type": "input_file", "file_id": file_record.id},
        ],
    }],
    previous_response_id=file_record.response_id,
)
upload_url expires after 3 minutes.

Reference