cerolabs 0.1.0

Python SDK for Cero Labs — agent-native expert escalation for AI workflows.

cerolabs

Python SDK for Cero Labs — the agent-native API for on-demand domain expertise.

When your agent hits an edge case, cerolabs routes the case to a verified domain expert and returns a structured, schema-validated response your workflow can use directly.

Install

pip install "git+https://github.com/os1123/cero-labs.git#subdirectory=sdk/python"

Requires Python 3.10+.

Package name for PyPI: cerolabs. The publish workflow is configured in this repo, but the live PyPI project still needs trusted-publisher setup.

Quick start

from cerolabs import Cero

cero = Cero(api_key="cero_...")

schema = {
    "type": "object",
    "properties": {
        "appeal_recommended": {"type": "boolean"},
        "reasoning": {"type": "string"},
    },
    "required": ["appeal_recommended", "reasoning"],
}

resolution = cero.escalate_and_wait(
    domain="healthcare.rcm",
    query="Should we appeal this CO-50 denial?",
    context={"cpt": "93015", "payer": "Blue Cross"},
    schema=schema,
)

print(resolution.answer["appeal_recommended"])
print(resolution.answer["reasoning"])

Pydantic models

If you already use Pydantic, pass the model class directly. The SDK will derive output_schema for Cero and validate the expert answer on the way back.

from pydantic import BaseModel

from cerolabs import Cero


class CodingDecision(BaseModel):
    cpt_codes: list[str]
    icd10_codes: list[str]
    notes: str | None = None


cero = Cero(api_key="cero_...")
resolution = cero.escalate_and_wait(
    domain="healthcare.coding",
    query="Verify the code set for this operative note.",
    context={"note_id": "OP-123"},
    schema=CodingDecision,
)

assert isinstance(resolution.answer, CodingDecision)
print(resolution.answer.cpt_codes)

Async client

import asyncio

from cerolabs import AsyncCero


async def main() -> None:
    async with AsyncCero(api_key="cero_...") as cero:
        resolution = await cero.escalate_and_wait(
            domain="legal.contract",
            query="Is this indemnity clause outside standard policy?",
        )
        print(resolution.answer)


asyncio.run(main())

API

• Cero(api_key, base_url?, timeout_s?): sync client
• AsyncCero(api_key, base_url?, timeout_s?): async client
• escalate(...): create an escalation and return immediately
• wait_for_resolution(escalation_id, ...): long-poll until resolved
• escalate_and_wait(...): convenience wrapper
• check_escalation(escalation_id): get current status
• cancel_escalation(escalation_id): cancel a pending/routed escalation
• submit_feedback(escalation_id, score, comment?): rate the expert answer
• list_escalations(...): list recent escalations
• list_domains(): live supported domains
• usage(): tenant credits, tier, and limits

Machine-readable docs

• Visual docs
• Agent docs (/docs.md)
• Agent discovery (/llms.txt)
• Extended agent guide (/llms-full.txt)

Publish locally

python -m pip install build twine
python -m build
python -m twine check dist/*

License

MIT