OCTOMIL / DOCUMENTATION

From a local test
to your first rollout.

Run your models on your users’ devices, with cloud fallback when needed. Start with one task and measure quality, latency, and cost before expanding.

Start with your workload

01 / EXPLORE

Run locally

Set up Octomil on macOS, Windows, or Linux and try a model on your hardware.

Get the CLI ↗02 / EVALUATE

Plan a pilot

Choose one feature, representative inputs, and a device mix. Set acceptance criteria before a rollout.

Follow the pilot guide →03 / INTEGRATE

Use the API

Explore authentication, model management, deployment operations, and telemetry.

Read the API reference →04 / CHOOSE

Explore models

Review model families and runtime options. Confirm compatibility with your platform and hardware.

Browse the catalog ↗

Choose your SDK

Each SDK has its own supported models and execution paths. An API call alone does not establish that inference ran locally. Follow the platform guide and verify where your workload executes.

Python

Local experiments and control-plane operations.

Python guide →

iOS

Apple app integration and Core ML execution paths.

iOS guide →

Android

Android integration, model delivery, and local runtimes.

Android guide →

Browser

Web integration with WebGPU and WASM runtime options.

Browser guide →

Measure before expanding

Compare real device output against your task requirements. Include model download size, memory, fallback, retries, and delivery costs when assessing the result. Savings estimates need your actual cloud rates and equivalent successful tasks.

Need help choosing the first workload? Book a conversation with Sean, or explore a pilot.

Source and support

Browse the source on GitHub. For deployment questions, contact team@octomil.com.