MindDuck doesn't sell access to AI. It works with models you connect with your own key (API key) from the service you choose, and you pay that service directly. Without a model MindDuck doesn't answer, but search, history import and documents work.
Four jobs for models
| Job | What it does | Where it's set |
|---|---|---|
| Chat model | Answers your messages. Can be changed per chat | Model for new chats, the model picker above the input field |
| Background model | Writes chat titles, summaries of long conversations and topics in the background | Models → Service |
| Speech recognition | Turns voice into text | Models → Dictation |
| Web search | Looks up fresh information on the internet | Models → Web search and the connection's Web access mode |
The easiest way to cover all four jobs is one OpenRouter key.
Background model
It decides how well MindDuck remembers long conversations: it compresses the older part of a conversation into a summary. Titles and topics help you find chats in search.
- Choose an inexpensive, fast model: a strong one isn't needed here, and there are many requests.
- Run tasks: Automatically runs tasks on their own; Only when requested makes MindDuck offer to run them each time.
- Right after your first connection, the chat model becomes the background model in Only when requested mode. Change it in the checklist (Settings → Overview → Background model) or in Models → Service.
Speech recognition
OpenRouter, OpenAI, Groq and Mistral AI have recognition models. Open Models → Dictation, choose a service and a model and turn on Dictate messages. More: Voice input.
Web search
Every connection has a Web access mode: Off or Automatic (recommended). In automatic mode MindDuck uses the model's built-in search if it has one. Otherwise a fallback search works: free Wikipedia by default, and for search across the whole internet connect Brave, Tavily or Exa in Models → Web search. More: Web search.
Which service to choose
We recommend OpenRouter. One key gives access to hundreds of models from different companies and covers all four jobs. Before you start, top up your OpenRouter balance and check its privacy settings: the link is in the connection form.
| In the Which service? list | When to choose it |
|---|---|
| Claude, ChatGPT, Gemini, Grok, Kimi | You already have a key for this service. All of them have built-in web search |
| DeepSeek, Mistral AI, Groq | You want this service's models directly |
| Ollama, LM Studio | The model runs on your computer, no key needed |
| Another service | Any service with an OpenAI-compatible API (an address with https://) |
You can connect several services and split the jobs between them.
How to connect
- Open Settings and expand the Model step. Or click Connect a model on the Where would you like to start? screen.
- In Which service? choose a service (OpenRouter - recommended by default).
- Click Where do I get a … key?, create a key on the service's site and copy it.
- Paste the key into API key and click the connect button. MindDuck checks the key and loads the list of models.
- In Model for new chats choose a model and click Start a chat.
You're done when Settings → Overview shows Ready to work. Then set up the other jobs on the Service, Dictation and Web search tabs of the Models screen.
A model on your own computer
Start Ollama or LM Studio and download a model in it. In MindDuck choose Ollama or LM Studio and leave the key empty. If the server runs on a different port, fix the Server address in Advanced settings. You can't connect to a model on your computer from a phone.
Keys and several connections
The key is stored encrypted and is sent only to its own service. With sync it moves to your other devices.
One connection is marked Default: it answers in new chats. Change it with the Use by default button on the Models screen.
What it costs
The service sets the prices; MindDuck adds nothing. Amounts in the app are approximate: the service bills the exact amount, including caching, reasoning, search and images. Check it in the service's own account.
Separate paid requests come from Test selected model, web search through the provider, image descriptions and the background model's work.
Messages when connecting
| Message | What to do |
|---|---|
| The provider rejected this API key | The key was mistyped or revoked. Create a new one |
| This API key does not have access to the provider or requested model | Check the key's permissions or your plan |
| The account has insufficient funds or the key spending limit was reached | Top up your balance or raise the key's limit |
| The provider rate limit or quota was reached | Wait or check your quotas |
| MindDuck could not reach the provider | Check your internet; for Ollama and LM Studio, check that the server is running |
| The provider connected, but returned no compatible chat models | For a local server, download a model |
If the Model step shows Connected, but no model is ready, expand the connection and click Check and reload models.
What's next
Updated October 7, 2026 · for version 0.2.12 · Found a mistake? Write to us