Documentation
AWSBee has three parts: a sizing engine that matches open models to NVIDIA GPU instances, a deployment manager that launches them as CloudFormation stacks in your own account, and a gateway that exposes every deployment through one OpenAI-compatible API.
Concepts
| Deployment | A named model endpoint. Its name is the model you pass to the API. It is either managed (a stack AWSBee launched in your account) or external (an OpenAI-compatible URL you registered). |
|---|---|
| Engine | vllm (open-source vLLM server, pinned version) or nim (NVIDIA NIM container, requires an NGC key). |
| API key | A bee_sk_… key that authenticates gateway calls. Shown once; revocable at any time. |
| Cloud account | The account AWSBee launches stacks into, connected with a scoped access key. |
Start here
- Quickstart — first request in minutes, with or without launching a GPU.
- Cloud account setup — IAM policy, GPU quotas and tokens.
- API reference — endpoints, streaming and errors.