---
name: theta-edgecloud
description: Work with Theta EdgeCloud, a GPU cloud for AI. Rent GPU nodes, run on-demand model inference, and read pricing or billing through its REST API. Use when a user mentions Theta EdgeCloud, TEC, Theta GPU rental, RTX or H200 GPUs on Theta, on-demand model APIs, or wants to deploy AI workloads on Theta.
metadata:
  website: https://www.thetaedgecloud.com
  docs-index: https://docs.thetatoken.org/llms.txt
  api-catalog: https://www.thetaedgecloud.com/.well-known/api-catalog
---

# Theta EdgeCloud

Theta EdgeCloud is a hybrid GPU cloud for AI workloads. It offers hosted data
center GPUs (for example H200), community RTX GPUs (5090, 4090, 3090 and
similar) at low hourly rates, on-demand model inference APIs, dedicated model
serving, Jupyter notebooks, and GPU clusters for training.

## Discover documentation

The documentation index is at https://docs.thetatoken.org/llms.txt. Append
`.md` to any documentation page URL to get its markdown version. Start with:

- Overview: https://docs.thetatoken.org/docs/theta-edgecloud-overview.md
- REST API: https://docs.thetatoken.org/docs/theta-edgecloud-api.md
- API keys: https://docs.thetatoken.org/docs/edgecloud-api-keys.md
- On-demand model APIs: https://docs.thetatoken.org/docs/edgecloud-on-demand-model-apis.md
- GPU nodes with SSH: https://docs.thetatoken.org/docs/edgecloud-ai-training-with-gpu-nodes.md
- Dedicated model serving: https://docs.thetatoken.org/docs/serving-generative-ai-models.md

The RFC 9727 API catalog at https://www.thetaedgecloud.com/.well-known/api-catalog
lists every public API with links to its documentation.

## Authentication

There is no OAuth or agent self-registration. Every API call needs a
project-scoped API key, created by a human in the dashboard:

1. Sign up or log in at https://www.thetaedgecloud.com/sign-up.
2. Create an API key for the project (see the API keys page above).
3. Send it in the `x-api-key` header on every request.

Never ask the user to paste an API key into chat. Ask them to set it in an
environment variable such as `TEC_API_KEY` and reference that.

## REST API

Base URL: `https://controller.thetaedgecloud.com`. All paths are versioned
under `/v1` and scoped to a project ID.

| Capability                   | Method and path                                                        |
| ---------------------------- | ---------------------------------------------------------------------- |
| List GPU resources           | `GET /v1/projects/{projectId}/gpu-resources`                           |
| Get a GPU resource           | `GET /v1/projects/{projectId}/gpu-resources/{resourceId}`              |
| List GPU node templates      | `GET /v1/projects/{projectId}/gpu-deployment-templates`                |
| List or create GPU nodes     | `GET, POST /v1/projects/{projectId}/gpu-deployments`                   |
| Get or delete a GPU node     | `GET, DELETE /v1/projects/{projectId}/gpu-deployments/{deploymentId}`  |
| Start or stop a GPU node     | `POST /v1/projects/{projectId}/gpu-deployments/{deploymentId}/start` or `/stop` |
| GPU node events or logs      | `GET /v1/projects/{projectId}/gpu-deployments/{deploymentId}/events` or `/logs` |
| Billing                      | `GET /v1/projects/{projectId}/billing/balance`, `/usage`, `/top-ups`, `/pricing` |

Successful responses use the envelope `{"status": "success", "body": {...}}`.

Example:

```bash
curl -sS \
  -H "x-api-key: ${TEC_API_KEY}" \
  "https://controller.thetaedgecloud.com/v1/projects/${TEC_PROJECT_ID}/gpu-resources"
```

## On-demand model inference

Serverless inference for hosted open models (LLMs, image, video, speech) is
billed per token or per unit of output with no GPU to manage. The model list,
pricing and request formats are in the on-demand model APIs documentation
linked above, and the live catalog is in the dashboard at
https://www.thetaedgecloud.com/dashboard/ai/service/on-demand-model-apis.

## MCP server

Theta EdgeCloud hosts an official MCP server at
`https://controller.thetaedgecloud.com/mcp` (Streamable HTTP, POST only).
Prefer it over hand-written HTTP calls when your client supports MCP.

- Connect with OAuth 2.1 (dynamic client registration and PKCE; discovery at
  `/.well-known/oauth-authorization-server` on that host), or send API keys
  directly: the on-demand key as `Authorization: Bearer <key>` and the project
  key as `x-api-key: <key>`.
- Tools follow the granted scopes: `inference` (list_services, infer,
  get_request_status, get_upload_url), `gpu:read` and `gpu:write` (GPU Node
  resources, templates, lifecycle, events, logs) and `billing:read` (balance,
  usage, top-ups, pricing). Mutating tools require `confirm: true` after the
  user approves the billable action.
- Server card: https://www.thetaedgecloud.com/.well-known/mcp/server-card.json
- Local stdio alternative for inference only: `npx @thetalabs/on-demand-api-mcp`
  (https://github.com/thetatoken/on-demand-api-mcp).

Call `list_services` first for inference; it returns each model's alias and an
example input. Call `list_gpu_resources` before creating a GPU Node and state
the hourly price to the user.

## Pricing

- Community RTX GPU rates and a fit quiz: https://www.thetaedgecloud.com/rtx-gpu
- Current featured GPU and inference rates: https://www.thetaedgecloud.com/
- Programmatic pricing: the billing `pricing` endpoint above.

## Safety rules

- Creating or starting a GPU node provisions billable capacity. Confirm with
  the user before doing so, and state the hourly rate.
- Stopping a GPU node keeps its configuration and can be restarted. Deleting
  a GPU node is permanent.
- Read-only calls (list, get, events, logs, billing) are safe to run freely.