API Reference
Donkey Framework
Build production-ready AI applications with modular, composable components.
⚡ Starter#
The quickest way to get up and running is with the starter bundle, which includes the core framework plus our most popular integrations:
$ pip install donkeyai
$ uv add donkeyai
🔧 Advanced (recommended for production)#
For production deployments, we recommend installing only the core package plus the specific integrations you actually use:
$ pip install donkey-core
$ pip install donkey-llms-*
$ pip install donkey-vector-stores-*
$ uv add donkey-core
$ uv add donkey-llms-*
$ uv add donkey-vector-stores-*
Installing individual packages instead of the full bundle keeps your environment lean:
- Smaller footprint — Fewer dependencies means smaller Docker images and faster cold starts.
- Fewer conflicts — Avoid pulling in transitive dependencies from integrations you never use.
- Security surface — Less third-party code to audit, patch, and monitor for vulnerabilities.
- Predictable upgrades — Control exactly which integrations get updated and when.
✨ Highlight features#
- Modular Architecture — Pick only the components you need. Each integration is its own installable package.
- Enterprise Integrations — First-class support for watsonx.ai, Elasticsearch, Chroma, Hugging Face, and more.
- Built-in Observability — Instrument your pipelines with OpenTelemetry-compatible tracing and custom metrics.
- Guardrails — Apply input/output guardrails to keep your AI applications safe and compliant.
- Async-First Workflows — Event-driven workflow engine with fan-out/fan-in, shared state, and built-in HTTP server.
👋 Contributing#
We welcome contributions! Please see our issue templates to get started.