Meta

The Meta API uses the OpenAI-compatible Responses API (POST /v1/responses) for the Muse Wonder and Muse Spark reasoning models. OpenClaw provides Meta as an official external plugin.

Getting started

openclaw plugins install @openclaw/meta-provider

openclaw onboard --auth-choice meta-api-key

openclaw onboard --non-interactive --accept-risk --skip-health \  --auth-choice meta-api-key \  --meta-api-key "$MODEL_API_KEY"

export MODEL_API_KEY=<key>

openclaw models list --provider meta

Non-interactive setup

openclaw onboard --non-interactive --accept-risk --skip-health \  --mode local \  --auth-choice meta-api-key \  --meta-api-key "$MODEL_API_KEY"

--mode defaults to local, so this is the same run as the Direct flag command above. Run it on the Gateway host: remote-client onboarding (--mode remote) only configures the local client connection and does not set up provider credentials on the server.

Built-in catalog

Prices and data-use terms come from Meta's pricing and rate limits documentation. Meta's model catalog positions Muse Wonder for complex coding and long-running agents, and Muse Spark for lower-cost workloads. New setups keep Muse Spark 1.3 as their default; Wonder is an explicit selection.

Muse Wonder is available in Standard mode only on Meta Model API. There is no muse-wonder-1.0-contributor API model. Muse Wonder Full (muse-wonder-full-1.0) is available only through the native Muse Code CLI on Power Usage and Maximum Usage subscriptions, not through this API provider. Setting Wonder's reasoning effort to max does not select Wonder Full. See Meta's availability table.

Capabilities:

• Text and image input through OpenClaw
• Tool calling and streaming
• Reasoning effort: minimal, low, medium, high, xhigh (OpenClaw default: high)
• max effort for Muse Wonder and Muse Spark 1.3 Standard; not for Contributor models
• Stateless encrypted reasoning replay (store: false, include: ["reasoning.encrypted_content"])

Meta's model catalog lists text, image, video, audio, and PDF input for these models. OpenClaw's model catalog directly represents text and image input only; the other upstream modalities are not model-manifest input values.

OpenClaw uses a 131,072-token catalog output cap for these models. Meta does not publish an exact maximum-output figure; this cap is not a claim about Meta's maximum. Output tokens include both reasoning and visible text.

OpenClaw explicitly selects high when no thinking level is configured. This is an OpenClaw default, not Meta's omitted-parameter behavior: Meta's reasoning documentation says that when reasoning.effort is omitted, the model reasons at a model-determined level.

Manual config

{  env: { vars: { MODEL_API_KEY: "<key>" } },  agents: {    defaults: {      model: { primary: "meta/muse-wonder-1.0" },      models: {        "meta/muse-wonder-1.0": { alias: "Muse Wonder" },      },    },  },}

Muse Code runtime

By default, meta/* models run on the built-in OpenClaw runtime through the Responses API. To run them through Meta's Muse Code CLI instead, install the @openclaw/muse plugin and set agentRuntime: { id: "muse" } on the model or provider. See Muse Code agent runtime.

In Settings → Models, choose Connect provider → Meta → Muse Code to sign in and approve native permissions. If an AI model is already available, this enables Muse Code without replacing your active model. When no model is available, the AI setup flow also selects the verified Muse model.

Smoke test

export MODEL_API_KEY=<key>pnpm test:live -- extensions/meta/meta.live.test.ts

The live suite exercises enabled Meta cases against POST /v1/responses. To test only Wonder, add -t muse-wonder-1.0. Its live cases cover text, image input, reasoning effort, and a tool-call round trip with encrypted reasoning replay.

Related