AI Providers
Study Agent works with local and cloud models. Ollama is the default — everything stays on your machine. Cloud providers are opt-in and need an API key.
| Provider | Local? | You need | Notes |
|---|---|---|---|
| Ollama (local) | Local | Ollama running locally, a pulled chat model | Default provider. Requests go to your configured URL — usually http://localhost:11434 — so nothing leaves your machine. |
| OpenAI | Cloud | OpenAI account + API key | Messages and prompt context are sent to OpenAI servers. |
| Anthropic Claude | Cloud | Anthropic account + API key | Messages and prompt context are sent to Anthropic servers. Extended thinking is not wired up. |
| OpenRouter | Cloud | OpenRouter account + API key | One key for many models. Messages are sent to OpenRouter and the selected model provider. |
| Groq | Cloud | Groq account + API key | Messages and prompt context are sent to Groq servers. |
| Gemini | Cloud | Google AI Studio API key | Messages are sent to Google. New keys start with `AQ.` — older keys may be blocked (HTTP 401). |
| Custom (OpenAI-compatible) | Depends | Any OpenAI-compatible server URL (e.g. LM Studio http://localhost:1234/v1) | Local if your server is local. Each provider keeps its own base URL, API key, and model. |
Ollama (default, local)
Install Ollama,
run ollama serve, pull a model, then point the plugin at
http://localhost:11434 and Refresh the model list. Nothing leaves
your machine. Thinking models additionally support
Enable thinking for reasoning before replying.
Cloud providers
Pick the provider in settings, paste your API key, Refresh, and select a model. Messages — plus the study-assistant identity, selected skill, custom instructions, and active-file context when enabled — are sent to that provider's servers. Per-provider protocol and privacy details: AI providers guide.
Custom (OpenAI-compatible)
Any server speaking the OpenAI chat API works — LM Studio
(http://localhost:1234/v1), vLLM, Ollama's OpenAI endpoint, and more.
Custom servers may block streaming via CORS; the plugin then automatically falls
back to a non-streaming request (which can't be cancelled once started).
Gemini keys
Google now issues new AQ. keys, and older keys may be blocked with
HTTP 401. If Gemini rejects your key, create a fresh one in Google AI Studio
and paste it again.
Each provider remembers its own setup
Base URL, API key, and model are saved per provider, so switching between local and cloud doesn't lose your local configuration.