Meta AI Integration Guide

VT Code has a first-class Meta AI provider for Meta's Muse models. Prefer this provider when you want to call Meta directly; use the OpenRouter guide when you specifically want OpenRouter billing, routing, or fallback options.

Prerequisites

Create a Meta AI API key using the Meta AI developer documentation. Meta's documentation uses MODEL_API_KEY; VT Code accepts that variable and also supports the provider-specific META_API_KEY variable.

export MODEL_API_KEY="your-meta-api-key"
# Or: export META_API_KEY="your-meta-api-key"

Quickstart

Use the official provider and its default Muse model:

vtcode --provider meta --model muse-spark-1.2 chat

The equivalent persistent configuration is:

[agent]
provider = "meta"
default_model = "muse-spark-1.2"

The endpoint defaults to https://api.meta.ai/v1. Set META_BASE_URL when a compatible gateway or test endpoint is required.

Curated models

Model	Notes
muse-spark-1.2	Default and latest curated Standard-tier Muse Spark model
muse-spark-1.1	Previous curated Standard-tier Muse Spark model
muse-spark-1.2-contributor	Opt-in Contributor-tier variant; review Meta's data-use terms before use

All three models are exposed through the /model picker and support streaming, function tools, structured output, multimodal input, and reasoning effort controls through Meta's OpenAI-compatible Chat Completions API.

See the Meta model catalog, Chat Completions protocol, and authentication guide for provider-side details.

Troubleshooting

Symptom	Resolution
Missing credentials	Set META_API_KEY or the documented MODEL_API_KEY variable.
Model rejected	Use one of muse-spark-1.1, muse-spark-1.2, or muse-spark-1.2-contributor for the official Meta provider.
Custom endpoint failure	Confirm the endpoint implements Meta's /v1/chat/completions protocol and set META_BASE_URL.