Ready-made tasks and shared services for AI jobs.
@metreeca/muse brings ready-made @metreeca/flow tasks for putting model work into a pipeline: prompting a model, extracting structure from free text, embedding content for retrieval. The tasks run under the @metreeca/gear job executor, which supplies the shared services they draw on.
- Ready-Made Tasks: prompting, extraction and embedding, chaining alongside any other task
- Shared Services: the model clients, credentials and response caches a run needs, built on demand and released as it ends
- Custom Bindings: a stubbed, throttled or recorded model swapped in for a run, leaving the job untouched
- Minimal Footprint: one package per model provider, each pulling in only the SDK that provider needs
Important
Pipelines are server-side workloads targeting Node.js 22 or later, relying on facilities such
as the filesystem, the process environment and fetch. The packages are not intended for the browser.
npm install @metreeca/gear # job executor and shared services
npm install @metreeca/muse # model access contracts and shared services
npm install @metreeca/muse-<provider> # task package, one per model providerWarning
TypeScript consumers must use "moduleResolution": "nodenext"/"node16"/"bundler" in tsconfig.json.
The legacy "node" resolver is not supported.
Install the core package, then add a task package for each model provider the pipeline reaches. Provider packages are self-contained leaves, each pulling in only the SDK its own provider needs. The job executor comes from @metreeca/gear, which the core package pulls in transitively; install it directly to set up and run a job.
| Package | Description |
|---|---|
| @metreeca/muse | AI tasks and shared services |
| @metreeca/muse-google | Google model connectors |
Note
Each package documents its own API in its README and API reference; for complete coverage, see the API reference.
- open an issue to report a problem or to suggest a new feature
- start a discussion to ask a how-to question or to share an idea
This project is licensed under the Apache 2.0 License – see LICENSE file for details.