opencode-swisscomgpt

SwisscomGPT provider for OpenCode - Access GPT-3.5, GPT-4, and GPT-5 via Swisscom's internal AI platform.

Features

• 🚀 Access to GPT-5 and other models via Swisscom's GPU cluster
• 🔐 JWT-based authentication via gpt.swisscom.com
• 🔄 Full streaming support (SSE)
• 🛠️ Easy integration with OpenCode

Installation

From npm (when published)

npm install opencode-swisscomgpt
# or
bun add opencode-swisscomgpt

From local source

# Clone and build
git clone https://github.com/mac-110/opencode-swisscomgpt.git
cd opencode-swisscomgpt
bun install
bun run build

# Install in OpenCode's config directory
cd ~/.config/opencode
bun add /path/to/opencode-swisscomgpt

Prerequisites

You need a Swisscom employee account to use this provider.

Getting your JWT Token

1. Log in to gpt.swisscom.com
2. Open Browser DevTools (F12)
3. Go to the Network tab
4. Reload the page or make any request
5. Find a fetch request and look at the Headers tab
6. Copy the token from the Authorization: Bearer <token> header

Configuration

1. Set Environment Variable

export SWISSCOMGPT_API_KEY="your-jwt-token"

# Or add to ~/.zshrc / ~/.bashrc for persistence
echo 'export SWISSCOMGPT_API_KEY="your-jwt-token"' >> ~/.zshrc

2. OpenCode Config

Add the provider to your ~/.config/opencode/opencode.json:

{
  "provider": {
    "swisscomgpt": {
      "npm": "opencode-swisscomgpt",
      "env": ["SWISSCOMGPT_API_KEY"],
      "options": {
        "baseURL": "https://api.rastro.icp.swisscom.com"
      },
      "models": {
        "gpt-5": {
          "name": "GPT-5",
          "id": "gpt-5",
          "attachment": true,
          "tool_call": true,
          "temperature": true,
          "cost": { "input": 0, "output": 0 },
          "limit": { "context": 128000, "output": 16384 },
          "modalities": { "input": ["text"], "output": ["text"] }
        },
        "gpt-5-instant": {
          "name": "GPT-5 Instant",
          "id": "gpt-5-instant",
          "attachment": true,
          "tool_call": true,
          "temperature": true,
          "cost": { "input": 0, "output": 0 },
          "limit": { "context": 128000, "output": 16384 },
          "modalities": { "input": ["text"], "output": ["text"] }
        },
        "gpt-5-thinking": {
          "name": "GPT-5 Thinking",
          "id": "gpt-5-thinking",
          "reasoning": true,
          "attachment": true,
          "tool_call": true,
          "temperature": true,
          "cost": { "input": 0, "output": 0 },
          "limit": { "context": 128000, "output": 32768 },
          "modalities": { "input": ["text"], "output": ["text"] }
        },
        "gpt-4-turbo": {
          "name": "GPT-4 Turbo",
          "id": "gpt-4-turbo",
          "attachment": true,
          "tool_call": true,
          "temperature": true,
          "cost": { "input": 0, "output": 0 },
          "limit": { "context": 128000, "output": 4096 },
          "modalities": { "input": ["text"], "output": ["text"] }
        }
      }
    }
  },
  "model": "swisscomgpt/gpt-5",
  "small_model": "swisscomgpt/gpt-4-turbo"
}

Important: OpenCode requires explicit model definitions in the models object. Each model you want to use must be defined with its capabilities and limits.

Model Configuration Reference

| Field | Type | Description | |-------|------|-------------| | name | string | Display name in OpenCode | | id | string | API model identifier | | tool_call | boolean | Supports function calling | | reasoning | boolean | Model has reasoning/thinking capability | | attachment | boolean | Supports file attachments | | temperature | boolean | Supports temperature parameter | | cost.input | number | Cost per 1M input tokens (USD) | | cost.output | number | Cost per 1M output tokens (USD) | | limit.context | number | Max context window size | | limit.output | number | Max output tokens | | modalities.input | array | Supported input types: text, image, audio | | modalities.output | array | Supported output types: text, image, audio |

Available Models

GPT-5 Models

| Model ID | Description | |----------|-------------| | gpt-5 | Default GPT-5 model | | gpt-5-instant | ⚡ Fastest replies, no deep thought | | gpt-5-thinking-mini | 💡 Light, efficient reasoning | | gpt-5-thinking | 🧠 Slower, more detailed reasoning |

GPT-4 Models (Legacy)

| Model ID | Description | |----------|-------------| | gpt-4o | Great for general tasks | | gpt-4o-mini | Faster for simple tasks | | gpt-4-turbo | High performance | | gpt-4.1 | Optimized for coding |

Reasoning Models

| Model ID | Description | |----------|-------------| | o3 | Advanced reasoning model | | o3-mini | Fast reasoning model |

Claude

| Model ID | Description | |----------|-------------| | claude-3.5-sonnet | Great for creative writing and code |

Swiss Models 🇨🇭

| Model ID | Description | |----------|-------------| | apertus | Swiss-trained open model | | llama-4-scout | Large context model |

Usage with OpenCode

# Verify models are available
opencode models swisscomgpt

# Start OpenCode (uses configured default model)
opencode

# Switch model in session
/model swisscomgpt/gpt-5-thinking

Programmatic Usage

You can also use this provider directly in your code:

import { createSwisscomGPT } from 'opencode-swisscomgpt';
import { generateText } from 'ai';

const swisscomgpt = createSwisscomGPT({
  apiKey: process.env.SWISSCOMGPT_API_KEY!,
});

const { text } = await generateText({
  model: swisscomgpt('gpt-5'),
  prompt: 'Explain quantum computing in simple terms.',
});

console.log(text);

With Streaming

import { createSwisscomGPT } from 'opencode-swisscomgpt';
import { streamText } from 'ai';

const swisscomgpt = createSwisscomGPT({
  apiKey: process.env.SWISSCOMGPT_API_KEY!,
});

const result = await streamText({
  model: swisscomgpt('gpt-5'),
  prompt: 'Write a short story about AI.',
});

for await (const chunk of result.textStream) {
  process.stdout.write(chunk);
}

API Reference

createSwisscomGPT(config)

Creates a SwisscomGPT provider instance compatible with Vercel AI SDK.

Config Options

| Option | Type | Default | Description | |--------|------|---------|-------------| | apiKey | string | required | JWT Bearer token from gpt.swisscom.com | | baseURL | string | https://api.rastro.icp.swisscom.com | API base URL | | threadId | string | auto-generated | UUID for conversation continuity | | policy | string | SKIP | API policy parameter |

How It Works

This provider wraps @ai-sdk/openai-compatible and transforms requests for SwisscomGPT's API:

1. URL Transformation: /v1/chat/completions → /v2/chat/completions/{threadId}?policy=SKIP
2. Authentication: Uses JWT Bearer token in Authorization header
3. Thread ID: Auto-generates UUID per request for conversation tracking

Troubleshooting

"Provider not found: swisscomgpt"

1. Make sure the package is installed in ~/.config/opencode/node_modules/
2. Verify your opencode.json has explicit model definitions
3. Check that the package is built (dist/ folder exists)

Token Expired

JWT tokens from gpt.swisscom.com expire after ~24h. Get a fresh token from the browser.

Network Errors

Make sure you're connected to the Swisscom network (VPN if remote).

Security Notes

• Never commit your JWT token to version control
• Use environment variables for the API key
• Tokens are tied to your personal Swisscom account

Contributing

Contributions welcome! Please read our Contributing Guide.

License

MIT © Swisscom / Alexander Schäfer

Made with ❤️ for Swisscom developers