Setup and models

The setup guide

Six short steps on first start: Welcome, Add a provider, Choose models, Choose your documents folder, Licence and privacy (where you pick the telemetry level), and Try it, which reviews a short sample document. Every step can be skipped and changed later in Settings; Help → Setup guide opens it again.

Providers

A provider is where the AI model runs. Add one in the setup guide or in Settings → Providers:

• Ollama or llama.cpp on this computer: private, nothing leaves your machine. The addresses are usually http://localhost:11434/v1 (Ollama) and http://localhost:8080/v1 (llama.cpp).
• Custom (OpenAI-compatible): a company gateway, a hosted model or a server on another computer: its address and, if it needs one, a key.
• Cloud providers: OpenAI, Anthropic, Google Gemini, DeepSeek, OpenRouter, NVIDIA. Paste your API key; the document text you review is sent to that provider under its terms.

Test (or Discover models) asks the provider for its model list. Keys are stored on your computer only.

Choosing models

The model selector (the model name at the right of the status bar, or Settings → Models) has three choices. For each, pick the provider first, then the model; a provider with many models shows a filter box: type part of a name.

• Reviewer drives the whole analysis: reading each section, finding issues and writing comments. Use your strongest model here.
• Verifier checks each finding and its quotes before it ships. It can be the same model, or a faster one.
• Vision describes the pictures in a document. Off by default; see Pictures.

Each change is saved at once.

Is my model good enough?

Settings → Models → Check this model runs five short tests on the Reviewer model (following instructions, quoting exactly, structured answers) and tells you whether its reviews will be useful. A model that fails gets a warning when you start an analysis.

Pictures in documents

When you choose a Vision model, DocReview describes each picture while it prepares a document, so the review can take diagrams and screenshots into account; the pictures go to that model. With Vision off, pictures are not described and none leaves your computer. Settings → Analysis → Advanced → Show the model the document's images controls whether the Reviewer itself is shown images (only models that can see them).

Free tiers

Free tiers of cloud providers may use what you send them for training, and they limit requests per minute and per day. DocReview warns you once before you use one, waits when a limit is reached and shows in the status bar when it will continue.

Word support (LibreOffice)

DocReview reads Word and Excel files with LibreOffice. The Windows installer includes it; on macOS and Linux install LibreOffice yourself. The setup guide shows a notice when it is missing or does not work.