Private AI inference
Run open AI models on GPUs you own
AWSBee launches open-weight models like Llama, Qwen and Mistral on NVIDIA GPUs in your own cloud account, then gives your team one simple API with usage and cost tracking. Your data stays where it is.
Free for up to 2 deployments · No credit card · Your cloud provider bills GPU hours directly
- Model
- Llama 3.1 8B Instruct
- GPU
- 1× NVIDIA L4
- Time to first token
- 180 ms
- Cost
- $0.22 / 1M tokens
Platform
Everything you need to run open models in production
Skip the weeks of GPU sizing, drivers, inference servers and networking. AWSBee handles the setup and keeps it running.
Your account, your data
Models run on GPU instances in your own cloud account. Prompts, outputs and weights never pass through a third-party model host.
Right-sized GPUs
Pick a model and a context length; the sizing engine works out memory for weights and KV cache and suggests the cheapest NVIDIA GPU that fits.
vLLM or NVIDIA NIM
Serve open-weight models with vLLM, or run NVIDIA NIM containers with your NGC key. Your application code doesn't change.
One OpenAI-compatible API
Every deployment sits behind one endpoint and one set of keys. Point any OpenAI SDK at it by changing the base URL.
Cost per token
Latency, time to first token, throughput and spend per million tokens for each deployment, so you know what a model really costs.
Secure by default
Model ports only accept traffic from the gateway, volumes are encrypted, and API keys are stored hashed. Delete a deployment and it is gone.
Use cases
What teams run on it
Anything you would use a hosted AI API for, when the data is too sensitive or the volume too high to send elsewhere.
- Answer support tickets with a model that never sends customer data outside your account
- Summarize contracts, reports and call notes that can't go to a public AI service
- Generate embeddings for internal search and retrieval
- Classify or tag large batches of documents at a fixed hourly cost
- Compare an open model with your current hosted API, side by side
- Give every internal team one approved, metered AI endpoint
How it works
From cloud account to working API in three steps
- 01
Connect your cloud account
Grant a narrowly scoped permission set we provide. We check your GPU quotas by region before anything is launched.
- 02
Choose a model
Pick from a curated catalog of open-weight models, or bring your own. We recommend the GPU and launch a reproducible, hardened stack.
- 03
Call it and watch the numbers
Use the endpoint from your app, test prompts in the playground, and track latency, usage and cost per token from day one.
Pricing
Start free, upgrade when you go to production
Full pricingDeveloper
For individuals evaluating self-hosted models.
$0forever
Team
For teams running models in production.
$49per month
Enterprise
For regulated workloads and large GPU fleets.
Custom
See AWSBee with your own model
Book a 45-minute call. We'll size the model you have in mind, estimate the GPU cost, and show a live deployment end to end.