Python SDK · v1 stable

Typed. Async-first. Built for production.

One typed Python client for every product on the platform. Pydantic v2 throughout, idempotency keys handled for you, evidence verification a single call — with a matching CLI for the moments code is overkill.

SDK reference See the flagship agent

radmah-quickstart.ipynbpython 3.11

cell · in[1]

1from radmah_sdk import Client2 3client = Client(api_key="rm_live_…")4 5# Synthesize from a CSV upload6ds  = await client.datasets.upload("./customers.csv")7job = await client.synthesize.run(8    dataset_id=ds.id,9    model="synthesize",10    rows=10_000,11    seed=42,12)13await job.wait()14 15bundle = await client.evidence.fetch(job.id)16assert bundle.verify()  # cryptographic hash, offline17print(job.metrics.qa_score)

stream · out[1]

▶ uploading customers.csv ………… 4.7 MB
dataset ds_4a7c81 · 20 cols · 482 931 rows
▶ POST /v1/synthesize/jobs 202
job_9b3df1 · estimate 38 s · 14 credits
▶ training the engine on 17 numeric / 3 cat …
epoch 12 / 20 loss 0.083
epoch 20 / 20 loss 0.041
▶ generating 10 000 rows ………… 1.6 s
✓ K-S gate 0.018 · χ² gate ok · constraints ok
▶ fetching evidence bundle …
✓ chain verified root a4f2…d801
>>> 0.9569

Type coverage: 100%
Python: 3.11+
Notebooks: 6
Wheel: signed

◆ Six guarantees

Six things every SDK call guarantees.

The same guarantees whether you're uploading a CSV or driving the autonomous agent. No surprises, no hidden retries, no untyped JSON blobs.

rule · 01

100% typed

Pydantic v2 models for every request and response. mypy-strict on the client. Your IDE knows every field.

rule · 02

Async-first

Every method is awaitable. Sync wrappers exist where you need them, but async is the primary surface.

rule · 03

Idempotency built in

Idempotent retries on POSTs out of the box; the SDK manages your Idempotency-Key for you.

rule · 04

Streaming evidence

First-class helpers to subscribe to evidence-chain events and verify the BLAKE3 hash chain offline.

rule · 05

Contract-test runner

Drop the contract runner into your CI to lock the SDK↔API contract — caught upstream, not in prod.

rule · 06

Matching CLI

Every SDK call has a rady CLI sibling for shell pipelines, one-off ops, and the interactive REPL. See /developer/cli.

◆ Quickstart

From pip install to a signed bundle in 30 seconds.

shell · zsh

pip install radmah-sdk radmah-cli
rady auth login                          # OS-keyring credential, one-time

# 30-second mock dataset via the SDK
python3 -c "
from radmah_sdk import RadMahClient
c = RadMahClient(api_key='rm_live_...')
job = c.submit_job_with_budget(
    kind='mock', engine='mock_fast',
    prompt='200 SaaS accounts MRR 50-5000',
    rows=200, seed=42, max_credits=1,
)
print(job.wait().to_dataframe().head())
"

# Or one-shot via the CLI
rady mock -p "200 SaaS accounts MRR 50-5000" -n 200 -o bundle.csv

# Verify any evidence bundle offline
rady evidence verify bundle.tar.zst

sdk · python · main.py✓ bundle sealed

import asyncio
from radmah import Client

async def main() -> None:
    async with Client() as rd:
        # Seal the contract for a tabular synthesis run.
        job = await rd.jobs.submit(
            contract={
                "dataset": "adult-income.csv",
                "rows": 50_000,
                "seed": 1847,
            },
            quality_gate={"floor": 0.85},
        )

        # Stream run events until the bundle is sealed.
        async for ev in rd.events.subscribe(job.id):
            print(ev.stage, ev.verdict, ev.elapsed_ms)

        # Offline-verify the sealed bundle.
        ok = await rd.evidence.verify(job.id)
        assert ok.chain_root == job.expected_root

asyncio.run(main())

rd.events.subscribe6 / 6

14:30:00.112contract.sealpass
sha 9c10ab · canonical-JSON
14:30:00.421engine.resolvepass
tabular · flagship
14:30:04.608generate.rowspass
50 000 · 48 cols · p99 0.8 s
14:30:06.328quality.gatepass
87.65 % · floor 85 % · ok
14:30:06.901physics.projectpass
41 corrections · 0 final
14:30:07.214evidence.sealdone
root a4f2…d801 · signed

assert ok.chain_root == job.expected_root ✓

◆ Reference notebooks

Six reference notebooks. One per product surface.

01-mock.ipynb

Generate a sealed mock dataset from a prompt.

02-synthesize.ipynb

Train it on a CSV upload, fetch the bundle.

03-agent.ipynb

Drive the autonomous data scientist end-to-end.

04-scada.ipynb

Spin up a Virtual SCADA run, capture telemetry.

05-evidence.ipynb

Verify any bundle, walk the BLAKE3 hash chain offline.

06-webhooks.ipynb

Receive signed events, validate signatures.

◆ Support

Language support, version policy.

Surface	Versions	Status
Python	3.11 – 3.13	stable
Async	asyncio, anyio	stable
Pydantic	v2.x	required
HTTP	httpx 0.27+	vendored
TypeScript SDK	Q3	in design
Go SDK	Q4	planned

One install. Every product surface.

Bring an OpenAPI test suite if you have one, drop the contract runner into your CI, and you can ship in a day.

SDK installation docs

terminal · install

$ pip install radmah-sdk

Collecting radmah-sdk

Downloading radmah_sdk-0.9.3-py3-none-any.whl (182 kB)

Collecting pydantic>=2.6

Collecting httpx>=0.27

Successfully installed radmah-sdk-0.9.3

$ rady --version

radmah-sdk 0.9.3 · python 3.12 · macOS/linux/windows