- Open modelsCall them. No deploy.
- Your modelsDeploy from Hugging Face.
- One endpointEvery model, one key.
- Pay per tokenNothing while idle.
Open models, large and small.
Call open models and the models you deploy from one endpoint.
from openai import OpenAI
client = OpenAI(
base_url="https://api.boardwalk.cloud/v1",
api_key="bw_live_…",
)
# Your model, deployed from Hugging Face.
response = client.chat.completions.create(
model="my-model",
messages=[{"role": "user", "content": "Hello"}],
)
print(response.choices[0].message.content)Deploy your own in three steps.
Same stack, same API, same per-token pricing.
Pick a model
Connect the exact Hugging Face revision you want to run.
Connect Hugging Faceacme/my-model@ c1899de2acme/support-lora@ 41f8a0b3Qwen/Qwen3-0.6B@ a1b2c3d4Deploy the qualified config
See the hardware, precision, context, and benchmark before you deploy.
Use the endpoint
Call it by name, same client and key.
client.chat.completions.create( model="my-model", messages=[...], )
Shared capacity. Fewer cold starts.
Open models, full fine-tunes, and LoRA adapters share warm GPU capacity. Active models stay loaded; idle models wake on demand.
- Shared
- Warm GPU capacity
- On demand
- Model loading
- $0
- While idle
Qwen/Qwen3-0.6Bopen modelacme/my-modelyoursteam/support-agentShared GPU poolOpen and custom modelsWarm workers
Security and honesty.
Private by default.
Prompt and completion text is never stored.
- Content retained
- None
- Stored per request
- Token counts
- API keys
- Hashed
Billing you can recount.
usage is what the engine processed — no rounding up.
- Billed dimensions
- Four
- Receipts
- Every request
- Recount with
/v1/tokenize
Nothing is unknown.
Every model publishes exactly what serves it.
- Weights
- Pinned commit
- Context
- Tested at the limit
- Parameters
- Honored or rejected