Deploy your first inference endpoint

In this guide:

Confirm your organization context and review available clusters.
Pick a model from the Inference Models catalog and a target cluster with enough capacity.
Deploy the endpoint, wait for Active status, then connect your application through the Acasia API.

Before you begin

Requirement	Description
Acasia account access	You can log in to the Acasia Developer Portal.
Correct organization context	You are operating inside the intended organization.
Active cluster	Your organization has at least one active cluster available.
Available resources	The cluster has enough GPU, CPU, RAM, and storage for the selected model.
Model access	The model is available in the Inference Models catalog.
Hugging Face token, if required	Some models may require authenticated Hugging Face access.

Inference endpoints, clusters, and API keys are scoped per organization. Use the organization switcher to confirm you’re in the right one before deploying.

Step 1 — Review your dashboard

Dashboard area	What to check
Clusters	Confirm that at least one cluster is available.
Cluster status	Confirm the cluster is marked Active.
GPU model	Confirm the cluster has the GPU type needed for your model.
Live models	Check whether models are already running on the cluster.

Step 2 — Inspect cluster resources

Resource	Why it matters
GPU count	Determines whether the model can run with the required accelerator capacity.
GPU model	Confirms hardware compatibility.
CPU cores	Supports runtime services, preprocessing, and orchestration.
RAM	Supports model loading and serving processes.
Storage	Stores model files, cache, logs, and temporary outputs.
Status	The cluster must be Active before deployment.

Step 3 — Open Inference Models

From the left navigation in the Developer Portal, select Inference Models.

Page element	Purpose
Create New Inference	Starts the deployment workflow.
Hugging Face Token	Adds or manages Hugging Face access for gated models.
Active tab	Shows running inference deployments.
Terminated tab	Shows stopped or removed deployments.

Step 4 — Choose a model

Select Create New Inference and pick a model from the catalog.

Consideration	Guidance
Task	Select a model that matches your use case — text, image, video, audio, or multimodal inference.
Model size	Larger models usually require more GPU memory and compute.
Access requirements	Some models may require a Hugging Face token.

If the model is gated, add a Hugging Face token from the Inference Models page before deployment. Without it, provisioning will fail when the runtime tries to pull model weights.

Step 5 — Choose a cluster and deploy

Select the target cluster

Pick an Active cluster that has enough GPU, CPU, RAM, and storage for the selected model.

Review the deployment summary

Confirm the model, cluster, and any optional parameters.

Click Deploy Endpoint

Acasia Cloud records the desired deployment state. The control plane then provisions the model on the target cluster.

Step 6 — Verify the endpoint

Return to the Active tab on Inference Models and watch the endpoint move through its lifecycle states.

State	Meaning
Active	The endpoint is running and available.
Provisioning	The endpoint is being created or prepared.
Failed	The deployment did not complete successfully.
Terminated	The endpoint has been stopped or removed.

Step 7 — Call the endpoint

curl --request POST \
  --url https://endpoint.acasia.ai/v1/completions \
  --header "Content-Type: application/json" \
  --header "Authorization: Bearer $ACASIA_API_KEY" \
  --data '{
    "model": "your-model-id",
    "prompt": "Summarize the safety risks in a warehouse aisle.",
    "max_tokens": 300
  }'

Endpoint paths, model identifiers, and payload formats vary by deployment. Always confirm against the endpoint detail view in the Developer Portal.

Pre-deployment checklist

Check	Confirmation
Organization context	You are in the correct organization.
Cluster status	The target cluster is active.
GPU capacity	The cluster has enough GPUs for the model.
Memory and storage	The cluster has enough RAM and storage.
Model selection	The model matches your use case.
Hugging Face access	Token is configured if the model requires gated access.

Summary

Confirm the correct organization context
Review available clusters
Open Inference Models → Create New Inference
Choose a model and an active cluster
Click Deploy Endpoint
Verify the endpoint appears under Active
Connect your application through the Acasia API

Next steps

Create an API key

Generate and store a credential for production use.

Connect with SSH

Open a secure shell to your cluster.

Connect your IDE

Develop against a remote GPU cluster.

​Before you begin

​Step 1 — Review your dashboard

​Step 2 — Inspect cluster resources

​Step 3 — Open Inference Models

​Step 4 — Choose a model

​Step 5 — Choose a cluster and deploy

​Step 6 — Verify the endpoint

​Step 7 — Call the endpoint

​Pre-deployment checklist

​Summary

​Next steps

Create an API key

Connect with SSH

Connect your IDE

Before you begin

Step 1 — Review your dashboard

Step 2 — Inspect cluster resources

Step 3 — Open Inference Models

Step 4 — Choose a model

Step 5 — Choose a cluster and deploy

Step 6 — Verify the endpoint

Step 7 — Call the endpoint

Pre-deployment checklist

Summary

Next steps