octomil-sdk 2.1.7

Octomil — serve, deploy, and observe ML models on edge devices

Octomil
Run ML models locally. Train them together.

---

Install

pip install octomil-sdk

Quick Start

Serve a model locally with an OpenAI-compatible API:

octomil serve gemma-1b

curl http://localhost:8080/v1/chat/completions \
  -d '{"model": "gemma-1b", "messages": [{"role": "user", "content": "Hello"}]}'

SDK Usage

from octomil import Octomil

client = Octomil(api_key="oct_...", org_id="org_123")

# Register a model
model = client.registry.ensure_model(name="sentiment", framework="pytorch")

# Gradual rollout
client.rollouts.create(model_id=model["id"], version="1.0.0", rollout_percentage=10)

# A/B test
client.experiments.create(name="v1-vs-v2", model_id=model["id"],
                          control_version="1.0.0", treatment_version="1.1.0")

Federated Learning

Train across devices without centralizing data:

from octomil import DeviceAuthClient, FederatedClient

auth = DeviceAuthClient(base_url="https://api.octomil.com", org_id="org_123",
                        device_identifier="runtime-001")
await auth.bootstrap(bootstrap_bearer_token=token)

client = FederatedClient(auth_token_provider=lambda: auth.get_access_token_sync(),
                         org_id="org_123")
client.register()

assignment = client.get_round_assignment()
if assignment:
    client.participate_in_round(round_id=assignment["round_id"],
                                local_train_fn=my_train_fn)

CLI

Command
octomil serve <model>	Local inference server (OpenAI-compatible)
octomil pull <model>	Download a model
octomil push <file>	Upload a model
octomil deploy <model>	Deploy to devices
octomil convert <file>	Convert to CoreML / TFLite
octomil check <file>	Validate a model
octomil scan <path>	Security scan
octomil benchmark <model>	Latency benchmarks
octomil login	Authenticate

Documentation

docs.octomil.com/sdks/python

Contributing

See CONTRIBUTING.md.

License

MIT