OpenCode Integration

OpenCode Integration

This guide explains how to configure OpenCode to work with your Cube AI instance.
OpenCode is an AI-powered code editor that can use the models hosted through Cube AI for code generation, editing, and general LLM assistance.

Prerequisites

Before you begin, ensure you have:

• A running Cube AI instance
• OpenCode installed locally
• A Cube AI user account with a Personal Access Token (PAT)
• A Cube AI domain with at least one coding-capable model pulled
  (for example: Qwen Coder models)

1. Create a Personal Access Token (PAT)

OpenCode uses a Personal Access Token (PAT) for authentication.

To create a PAT:

1. Log in to the Cube AI UI
2. Click your profile avatar
3. Navigate to Profile → Personal Access Tokens
4. Click Create Token
5. Copy and store the token securely

PATs are long-lived and are the recommended authentication method for all integrations.

2. Pull Recommended Coding Models

Before using OpenCode, make sure coding models are available in your Cube AI domain.

Recommended models:

• qwen2.5-coder:7b-instruct-q4_K_M
• qwen2.5-coder:3b
• deepseek-coder:6.7b-instruct-q4_K_M

Pull a model using:

curl -k -X POST https://<cube-ai-instance>/proxy/<your_domain_id>/api/pull \
  -H "Authorization: Bearer <pat>" \
  -d '{
    "name": "qwen2.5-coder:7b-instruct-q4_K_M"
  }'

3. Verify Available Models

To list all models available in your domain:

curl -k https://<cube-ai-instance>/proxy/<your_domain_id>/v1/models \
  -H "Authorization: Bearer <pat>"

You should see the models you pulled earlier.

4. Install OpenCode

Follow installation instructions from the official guide:

OpenCode

Once installed, OpenCode will create its configuration folder:

~/.config/OpenCode/

5. Configure OpenCode to Use Cube AI

Edit or create the configuration file:

~/.config/OpenCode/config.json

Add:

{
  "$schema": "https://opencode.ai/config.json",
  "model": "cube/qwen2.5-coder:7b-instruct-q4_K_M",
  "provider": {
    "cube": {
      "npm": "@ai-sdk/openai-compatible",
      "name": "Cube AI",
      "options": {
        "baseURL": "https://<cube-ai-instance>/proxy/<your_domain_id>/v1",
        "apiKey": "<pat>"
      },
      "models": {
        "qwen2.5-coder:7b-instruct-q4_K_M": {
          "name": "Qwen 2.5 Coder 7B - Main",
          "tool_call": false,
          "reasoning": false
        }
      }
    }
  }
}

Replace

• <cube-ai-instance> → usually localhost
• <your_domain_id> → the domain ID
• <pat> → your Cube AI Personal Access Token

Important

Use HTTPS (https://...) unless Cube AI is explicitly running without TLS.

6. Verify OpenCode Connection

In the OpenCode editor, run:

opencode /models

You should see your Cube AI models listed.

Try a simple prompt to verify everything works.

Troubleshooting

❌ Authentication error

• Ensure the PAT is valid
• Verify "Authorization: Bearer <pat>" is sent internally
• Confirm the token belongs to a user with access to the domain

❌ Connection issues

• Ensure Cube AI is running (make up)
• Verify HTTPS/TLS configuration
• Check firewall rules

Summary

You have successfully connected OpenCode to your Cube AI instance using Personal Access Tokens (PATs).

This setup enables secure, private, and domain-isolated coding assistance powered by Cube AI models.