What is meta-model-cookbook?

This repository is a cookbook of example code for developers building agents and coding tools on top of the Meta Model API, which serves an AI model called Muse Spark. The examples come as Colab notebooks and standalone Python files, and each one is meant to be copied and run as a starting point rather than a full application. The main selling point is compatibility: the Meta Model API is designed as a drop-in replacement for the OpenAI SDK and the Anthropic SDK, and it also works with agent command-line tools like OpenCode and Claude Code. In practice this means you can point an existing OpenAI-style client at a new base URL, supply an API key, and keep the rest of your code unchanged. The default model, Muse Spark, has a context window of just over one million tokens, and a preview version is currently free to use. The recipes are organized into three sections. The first, API fundamentals, covers the basic building blocks: chat completions, streaming responses, tool and function calling, structured JSON output, prompt caching, reasoning tokens, vision input, long context handling, error handling with retries, and grounding responses in live web search. The second section, agent patterns, shows how to wire a model into an actual agent loop, including interleaved reasoning and tool use, managing context across a long-running task, validated code edits, and a copilot for triaging noisy alerts. The third section, use cases, walks through complete applications: reading data out of charts, fixing bugs from a screenshot, a hands-free assistant for smart glasses, generating slide decks, building and verifying websites and browser games with a coding agent, running model-generated code in a sandbox, orchestrating a team of four specialized agents on a shared task board, and driving a real desktop or Mac purely from screenshots. To get started, a reader needs a Meta Model API account and key, which they store as an environment variable before running any of the notebooks or scripts. License details are included in the repository's LICENSE file.