Adds Muse Wonder and Muse Spark models to the Copilot Chat model picker. Uses your Meta Model API key (BYOK).
Features
- Model
muse-wonder-1.0 (1,048,576-token context) — preview model for frontier coding and long-running agent work; supports max reasoning
- Model
muse-spark-1.3 (1,048,576 context, 131,072 max output) — latest Spark checkpoint, tuned for agentic workflows
- Model
muse-spark-1.3-contributor — same checkpoint, heavily discounted Contributor tier
- Model
muse-spark-1.2 / muse-spark-1.1 — earlier checkpoints, still supported
- Text and image input through this extension; native image parts are sent as base64
image_url content (no proxy)
- Reasoning effort:
minimal, low, medium (default), high, xhigh, and max (Muse Wonder and Standard-tier muse-spark-1.3)
- Works with Copilot agent mode, tools, instructions, MCP, and skills via
LanguageModelChatProvider
Meta's API supports additional Wonder/Spark modalities such as video, audio, and PDF, but this extension currently forwards text and image parts only. Muse Wonder Full is available through eligible Muse Code subscriptions, not through Meta Model API, so it cannot be selected in this extension.
Requirements
- VS Code 1.116+
- GitHub Copilot (Free, Pro, or Enterprise)
- Meta API key from dev.meta.ai (
LLM|...)
Install
- Install from the VS Code Marketplace or Open VSX
- Command Palette → Meta Spark: Set API Key
- Open Copilot Chat and select Muse Wonder or a Muse Spark model
Configuration
| Setting |
Default |
Description |
meta-spark-copilot.baseUrl |
https://api.meta.ai/v1 |
API base URL |
meta-spark-copilot.maxCompletionTokens |
0 |
Max output tokens (0 = API default) |
meta-spark-copilot.modelIdOverrides |
official IDs |
Override model IDs for proxies |
meta-spark-copilot.debugMode |
minimal |
minimal / metadata / verbose |
meta-spark-copilot.experimental.stabilizeToolList |
false |
Experimental tool-list stabilization |
Pricing and limits
Muse Wonder (muse-wonder-1.0, Standard mode only):
- Pricing: $4.00 / 1M input, $0.20 / 1M cached input, $20.00 / 1M output
- Rate limits: 3,000 RPM / 4M TPM per team
- Context window: 1,048,576 tokens shared between input and output. Meta does not publish a separate numeric maximum output-token limit; the API applies its model-dependent limit.
- Standard mode content is not used to train Meta models. Wonder is in preview; API availability may vary by country.
Muse Spark Standard tier (muse-spark-1.1, muse-spark-1.2, muse-spark-1.3):
- Pricing: $1.25 / 1M input, $0.15 / 1M cached input, $4.25 / 1M output
- Rate limits: 3,000 RPM / 4M TPM per team
- The
max reasoning effort is available on muse-spark-1.3 (Standard) only
Muse Spark Contributor tier (muse-spark-1.2-contributor, muse-spark-1.3-contributor):
- Pricing: $0.10 / 1M input, $0.002 / 1M cached input, $0.20 / 1M output
- Rate limits: 150 RPM / 3M TPM per team
- Your prompts and completions may be used to train future Meta models
See Meta pricing and rate limits and model availability.
Common errors
| Code |
Meaning |
401 invalid_api_key |
Check key format (LLM\|...) |
429 rate_limit_exceeded |
Wait for Retry-After |
400 content_policy_violation |
Content blocked by policy |
503 / 504 |
Transient server/gateway issue; retry |
Development
npm install
npm run compile
Press F5 to launch the Extension Host.
To build a distributable .vsix package:
npm run package
The output lands in dist/.
Building and versioning
- Compile:
npm run compile (clean + tsc). Watch mode: npm run watch.
- Lint / format:
npm run lint (oxlint) and npm run format (oxfmt).
- Package:
npm run package produces dist/meta-spark-for-copilot-<version>.vsix using @vscode/vsce. It runs vscode:prepublish first, which prepares the marketplace README (scripts/prepare-marketplace-readme.sh).
- Local install: from the Extension Host, run
Extensions: Install from VSIX... and pick the .vsix.
- Marketplace publish:
npm exec -- vsce publish --packagePath dist/<file>.vsix (requires VSCE_PAT), or npm exec -- ovsx publish <file>.vsix for Open VSX.
Versioning (SemVer):
- Versioning is manual. The extension version lives in
package.json (version). See docs/RELEASE.md for the full guide.
feat:-sized additions (new model checkpoints, new tiers, new capabilities) bump the minor version. Breaking changes to settings or model IDs bump the major version. fix:-sized changes bump the patch version.
- This release is 2.2.0 (minor): it adds Muse Wonder 1.0 without breaking existing model IDs or settings.
- To release: bump
version in package.json, add a CHANGELOG.md entry, build and test (npm run package), then publish with vsce / ovsx and tag the release on GitHub.
License
MIT
| |