Documentation

Connect your tools to a GPU Cloud deployment. Every deployment serves an OpenAI Compatible endpoint and an Anthropic-compatible Messages endpoint.

Getting started

  1. 1

    Create an account

    Sign up with your email address. Create account

  2. 2

    Top up your wallet

    Deployments are paid from a prepaid wallet in BDT. Top-ups are credited after manual verification.

  3. 3

    Deploy a model

    Choose a plan and a model. The dedicated GPUs are reserved for the term you pay for.

  4. 4

    Copy your API key

    Open the deployment page, copy the key and click Show to reveal a key later. Keys are secret: never commit them to source control.

Connect your tools

Replace <your-deployment> with the slug shown on your deployment page, <your-model-name> with its model name, and YOUR_API_KEY with your key. One endpoint can serve several models; send the name of the model you want in the model field.

Connection values for any tool
Setting Value Action
OpenAI base URL https://models.betopiacloud.com/c/<your-deployment>/v1
Anthropic base URL https://models.betopiacloud.com/c/<your-deployment>
Model name <your-model-name>

Replace YOUR_API_KEY with a key from your deployment page.

Coding agents (Claude Code, Codex, Cline, OpenCode) need a tool-calling model and a large context. Recommended: Qwen3.6-35B-A3B (tested with Claude Code and Codex) or Qwen3-Coder-30B, with a context size of 131,072 tokens (minimum 65,536). Change both in the Models card of your deployment page.

Claude Code talks to the Anthropic Messages API, which your deployment serves natively.

Install

npm install -g @anthropic-ai/claude-code

macOS / Linux

export ANTHROPIC_BASE_URL="https://models.betopiacloud.com/c/<your-deployment>"
export ANTHROPIC_AUTH_TOKEN="YOUR_API_KEY"
export ANTHROPIC_MODEL="<your-model-name>"
export ANTHROPIC_DEFAULT_SONNET_MODEL="<your-model-name>"
export ANTHROPIC_DEFAULT_OPUS_MODEL="<your-model-name>"
export ANTHROPIC_DEFAULT_HAIKU_MODEL="<your-model-name>"
export CLAUDE_CODE_MAX_CONTEXT_TOKENS=65536
export CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC=1
claude

Windows PowerShell

$env:ANTHROPIC_BASE_URL="https://models.betopiacloud.com/c/<your-deployment>"
$env:ANTHROPIC_AUTH_TOKEN="YOUR_API_KEY"
$env:ANTHROPIC_MODEL="<your-model-name>"
$env:ANTHROPIC_DEFAULT_SONNET_MODEL="<your-model-name>"
$env:ANTHROPIC_DEFAULT_OPUS_MODEL="<your-model-name>"
$env:ANTHROPIC_DEFAULT_HAIKU_MODEL="<your-model-name>"
$env:CLAUDE_CODE_MAX_CONTEXT_TOKENS="65536"
$env:CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC="1"
claude

Persistent: ~/.claude/settings.json

{
  "env": {
    "ANTHROPIC_BASE_URL": "https://models.betopiacloud.com/c/<your-deployment>",
    "ANTHROPIC_AUTH_TOKEN": "YOUR_API_KEY",
    "ANTHROPIC_MODEL": "<your-model-name>",
    "ANTHROPIC_DEFAULT_SONNET_MODEL": "<your-model-name>",
    "ANTHROPIC_DEFAULT_OPUS_MODEL": "<your-model-name>",
    "ANTHROPIC_DEFAULT_HAIKU_MODEL": "<your-model-name>",
    "CLAUDE_CODE_MAX_CONTEXT_TOKENS": "65536",
    "CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC": "1"
  }
}

Codex CLI uses the OpenAI Responses API, which your deployment serves under the /v1 base URL.

Install

npm install -g @openai/codex

Configuration: ~/.codex/config.toml

model = "<your-model-name>"
model_provider = "betopia"

[model_providers.betopia]
name = "Betopia GPU Cloud"
base_url = "https://models.betopiacloud.com/c/<your-deployment>/v1"
env_key = "BETOPIA_API_KEY"
wire_api = "responses"

Set your key (macOS / Linux)

export BETOPIA_API_KEY="YOUR_API_KEY"

Set your key (Windows PowerShell)

$env:BETOPIA_API_KEY="YOUR_API_KEY"

Then run codex.

Troubleshooting

401 Unauthorized

The key is wrong or has been revoked. Copy a current key from the deployment page, or create a new one.

503 Service Unavailable

The deployment is not running or has expired. Check its status and renew it if needed.

The first response is slow

The model is still loading onto the GPUs. Wait a moment and retry.

"context length exceeded"

The prompt is larger than the context size. Raise the context size in the Models card of your deployment page.

The agent cannot call tools

Use a tool-calling model such as Qwen3-Coder with a context size of at least 65,536 tokens.