# API Access uniGPT provides an **OpenAI-compatible API** that allows researchers and developers to integrate AI models into their own applications, scripts, and workflows. The API is available at: ``` https://gpt.uni-muenster.de/v1 ``` ## Getting an API Key To use the API, you need an API key (starting with `sk-...`). API keys are currently available for two use cases: ### Research Projects API keys for research projects are issued upon request. Contact the uniGPT team via our support channel: [**uniGPT Support on Mattermost**]() Please briefly describe your project and intended usage so we can set appropriate quotas. ### AI-Assisted Coding (OpenCode, Roo Code, etc.) API keys for use with coding assistants are available through a **self-service portal** at **[https://gpt-api.uni-muenster.de]()**, where you can generate your own key without manual approval. The portal is open to all employees. A small guide for OpenCode and VSCode is shown [below](<#using-coding-assistants>). Currently the following models are enabled for all employees via the portal: - **DeepSeek-v4-flash** - **Qwen3.8-27B** Additional models are enabled on request — contact the team via the support channel. If you generated your key before a model was added, you may need to **regenerate your key** to pick it up. > [!WARNING] > > **Beta** > > The self-service portal is currently in **beta**. Access and functionality may change. Stay tuned for announcements in the support channel. ## Using the Python OpenAI Client Install the OpenAI Python library: ``` pip install openai ``` ### Basic Example ``` from openai import OpenAI client = OpenAI( api_key="sk-YOUR_API_KEY", base_url="https://gpt.uni-muenster.de/v1" ) completion = client.chat.completions.create( model="gemma-4-31B-it", messages=[ {"role": "user", "content": "What is the capital of France?"} ] ) print(completion.choices[0].message.content) ``` ### Streaming Example ``` stream = client.chat.completions.create( model="gemma-4-31B-it", messages=[{"role": "user", "content": "Explain quantum computing briefly."}], stream=True, ) for chunk in stream: print(chunk.choices[0].delta.content or "", end="") ``` ## Using curl ``` curl https://gpt.uni-muenster.de/v1/chat/completions \ -H "Content-Type: application/json" \ -H "Authorization: Bearer sk-YOUR_API_KEY" \ -d '{ "model": "gemma-4-31B-it", "messages": [{"role": "user", "content": "Hello!"}] }' ``` ## Checking Your Key & Quotas You can inspect your API key and its quotas using the `/key/info` endpoint: ``` curl https://gpt.uni-muenster.de/key/info \ -H "Content-Type: application/json" \ -H "Authorization: Bearer sk-ENTER-KEY-HERE" ``` This returns information about your key and the configured quotas/limits. ## Limits, Concurrency & Caching - **Parallel requests**: By default, each API key allows **4 parallel requests**. Contact the team if you need more. - **Input caching**: Prompt caching is enabled, but the discount is **not yet billed** — you currently pay the full input price even for cached tokens. - **Reasoning**: Reasoning is returned via the OpenAI Chat Completions format as the `reasoning_content` field alongside `content`. > [!NOTE] > > **Limits are separate from the chat balance** > > The API rate limits are managed independently of the token balance shown in the web chat interface. Your API key's quotas are queryable via `/key/info`. ## Available Models via API You can use any of the **internal** models listed on the [Available Models]() page by specifying their name in the `model` parameter, including specialized models such as `medgemma-1.5-4b-it` for medical text and image tasks. Through the self-service portal, **DeepSeek-v4-flash** and **Qwen3.8-27B** are currently enabled for all employees; additional models are enabled on request. > [!WARNING] > > **Model access** > > Access to a model is granted per team/group. If you receive the error `team not allowed to access model`, the requested model is not enabled for your account. > [!TIP] > > **Drop-in replacement** > > Since the API is OpenAI-compatible, you can use it as a drop-in replacement in any tool or library that supports the OpenAI API — just change the `base_url` and `api_key`. ## Using Coding Assistants You can connect uniGPT to your ide and coding agents via the OpenAI-compatible API. You will need an [API key](<#getting-an-api-key>) from the self-service portal. ### Visual Studio Code Since version 1.104, Visual Studio Code can connect to uniGPT in the **built-in Chat** view via a custom provider. No extra extension is required. This uses VS Code's [custom endpoint]() provider. Chat works with your uniGPT API key and does **not** require a GitHub Copilot plan. 1. Open the Command Palette (`Ctrl+Shift+P` / `Cmd+Shift+P`) and run **Chat: Manage Language Models**. 2. Select **Add Models**, then **Custom Endpoint**. 3. Enter a group name such as `uniGPT` and paste your uniGPT API key when prompted. 4. Select **Chat Completions** as the API type. 5. VS Code opens the `chatLanguageModels.json` file. Fill in the model fields as shown below and save it. **Do not replace** the `apiKey` value. Keep the `${input:...}` variable that VS Code generated. ``` [ { "name": "uniGPT", "vendor": "customendpoint", "apiKey": "${input:chat.lm.secret.xxxxxxxxx}", "apiType": "chat-completions", "models": [ { "id": "DeepSeek-V4-Flash", "name": "uniGPT DeepSeek V4 Flash", "url": "https://gpt.uni-muenster.de/v1/chat/completions", "toolCalling": true, "vision": true, "maxInputTokens": 128000, "maxOutputTokens": 8000 }, { "id": "Qwen3.8-27B", "name": "uniGPT Qwen3.8 27B", "url": "https://gpt.uni-muenster.de/v1/chat/completions", "toolCalling": true, "vision": false, "maxInputTokens": 128000, "maxOutputTokens": 8000 } ] } ] ``` Set `id` to an **internal** model name from the [Available Models]() page. You can add further objects to the `models` array to switch between them in the picker. Use `maxInputTokens` **128000** and `maxOutputTokens` **8000**. For a multimodal model such as DeepSeek V4 Flash, set `"vision": true`. If the model does not appear in the picker immediately, restart VS Code. ### OpenCode OpenCode can use uniGPT as a [custom OpenAI-compatible provider](). You need an [API key](<#getting-an-api-key>) from the self-service portal. 1. Start OpenCode and run **`/connect`**. 2. Select **Other**, then enter `unigpt` as the provider ID. 3. Paste your uniGPT API key when prompted. OpenCode stores it in its auth file. Do **not** put the key in `opencode.json`. 4. Create or update your OpenCode config (`~/.config/opencode/opencode.json` for all projects, or `opencode.json` in a project root): ``` { "$schema": "https://opencode.ai/config.json", "model": "unigpt/DeepSeek-V4-Flash", "provider": { "unigpt": { "npm": "@ai-sdk/openai-compatible", "name": "uniGPT", "options": { "baseURL": "https://gpt.uni-muenster.de/v1" }, "models": { "DeepSeek-V4-Flash": { "name": "uniGPT DeepSeek V4 Flash", "limit": { "context": 128000, "output": 8000 } }, "Qwen3.8-27B": { "name": "uniGPT Qwen3.8 27B", "limit": { "context": 128000, "output": 8000 } } } } } } ``` The provider ID in `/connect` must match the key in `provider` (`unigpt`). Use `@ai-sdk/openai-compatible` because uniGPT speaks the Chat Completions API. If you already have an `opencode.json`, merge the `provider.unigpt` block into it rather than replacing the whole file. 1. Run **`/models`** and select **uniGPT DeepSeek V4 Flash**.