One kernel, every provider.

Frontier model API behind Claude

The @namzu/anthropic package wires Anthropic's Messages API into Namzu's LLMProvider interface. Tool calling, streaming, structured outputs, and stop sequences are mapped one-to-one with no abstraction tax.

GPT and o-series model API

@namzu/openai targets the OpenAI Chat Completions and Responses APIs. Function calling is mapped to the kernel's tool primitive; streaming, JSON mode, and structured outputs all pass through.

Multi-model unified gateway

@namzu/openrouter routes through OpenRouter, the unified gateway over Anthropic, OpenAI, Google, Mistral, Meta, and dozens of open-weight models. Switching models is a config change, not a code change.

Managed model API on AWS

@namzu/bedrock authenticates with AWS Bedrock via standard SigV4 and exposes the same kernel-level tool, streaming, and vision primitives as every other provider package.

Local model runner

@namzu/ollama talks to a local Ollama daemon over the standard HTTP API. Nothing leaves the machine; the kernel's sandboxing and scheduling primitives still apply, just without a network hop.

Desktop local model server

@namzu/lmstudio targets LM Studio's OpenAI-compatible local server. Identical surface to the OpenAI provider, pointed at localhost.

Generic OpenAI-compatible endpoint

@namzu/http is the escape hatch. Point it at any OpenAI-compatible HTTP endpoint — vLLM, Together, Fireworks, Groq, your own gateway, an OpenAI-compatible router — and Namzu treats it like a first-class provider.

Sandboxed desktop control

@namzu/computer-use exposes desktop control — screenshots, mouse, keyboard, file system — as a Namzu tool. The agent gets a virtual display; the kernel runs the sandbox; the host machine is never touched directly.