AI agents
AI provider
The AI provider is the one connection OmniGet uses for its built-in AI features: summarizing a video, the AI video tools, transcribing audio and a few helpers in the Superpowers. It is Off until you choose a provider, so nothing is sent to an AI service unless you set it up.
Where it is
Settings › AI and agents › AI provider. The Configure tab of LLM opens the same section.
The coding agents in Threads and the agents of the Chat use their own connections, set in LLM › Configure › Connections (Connections, models and limits). This page is only about the shared provider for the built-in features.
How to use it
- Open Settings › AI and agents.
- In Provider, choose a service. The list is in the table below.
- In Model, type the model identifier for that service. The field shows an example for each provider.
- Paste your key in Connection API key (or Anthropic API key when the provider is Anthropic). Ollama runs on your computer and asks for no key.
- For Ollama (local), New API / One API (relay) and Custom endpoint (local or remote), fill in Endpoint URL.
- Click Save and test. OmniGet saves the settings, sends one small request and shows AI connection works or the error from the provider.
The line under the buttons then reads Connection verified, Connection test failed or Not tested.
To turn AI off again, set Provider to Off and click Save. Saved keys stay on your computer; to delete one, select its provider, click Remove saved key and save.
Providers
| Provider in the list | Key | Endpoint |
|---|---|---|
| OpenAI | Required | Fixed |
| Anthropic | Required (Anthropic API key) | Fixed |
| OpenRouter | Required | Fixed |
| DeepSeek | Required | Fixed |
| Google Gemini | Required | Fixed |
| Groq | Required | Fixed |
| xAI (Grok) | Required | Fixed |
| Mistral | Required | Fixed |
| SiliconFlow | Required | Fixed |
| Requesty | Required | Fixed |
| New API / One API (relay) | Required | Your relay’s address |
| Ollama (local) | None | Prefilled with the local Ollama address |
| Custom endpoint (local or remote) | Optional | Any OpenAI-compatible URL |
Options
| Option | What it does | Default |
|---|---|---|
| Provider | Where AI requests go. | Off |
| Model | The model identifier for the chosen provider. | Empty (the field shows an example) |
| Connection API key / Anthropic API key | Your key. It is stored locally and never logged. Leave the field blank to keep a saved key. | No key |
| Endpoint URL | The OpenAI-compatible service URL. The hint below it says whether the destination is this computer (a loopback address) or a network endpoint. | Empty, except for Ollama |
| Save | Saves without testing. | |
| Save and test / Test connection | Saves any change, then sends one test request. | |
| Discard changes | Puts back the saved settings. |
Unsaved changes are kept while you move between Settings sections, as long as the app stays open.
What uses it
- Summarize a video in LLM › Work.
- The AI video tools of Downloads.
- Audio transcription, for example in Telegram.
- The AI helpers inside some Superpowers, such as Read and the League coach.
Common problems
“Configure an AI provider first”
The feature you opened needs the AI provider and it is still Off, or no model is saved. Pick a provider, type a model and save.
Transcription fails with Anthropic
Audio transcription uses the OpenAI-style transcription endpoint. It works with OpenAI and with OpenAI-compatible endpoints that offer transcription. It does not work with Anthropic.
“AI test failed”
The message after the colon comes from the provider. Check the model name first (it must be one your account can use), then the key, then the Endpoint URL.
Requests leave my computer
Only when the provider is a remote service. With Ollama (local), or a custom endpoint on a loopback address, the requests stay on your computer. See Privacy in the app.