AI providers (OpenAI, Gemini, Ollama) guide

AI providers are the three services Runova can call directly for a quick answer: OpenAI, Gemini and Ollama. You give Runova an API key (OpenAI, Gemini) or the address of a server (Ollama), pick a default model, and the features that need a model use it. Nothing here needs an AI CLI to be installed. Without a provider, Ask AI, Prompt Builder and the Help page’s Ask the guide chat have nothing to talk to.

Where to find it

Click Settings in the sidebar, then open the AI tab. The first card is Providers (“Keys and models for Ask AI, Prompt Builder and the other AI features”). It has two parts: Prompt Builder writes with at the top, a row of tiles that picks what writes your prompts, and Connections below it, with a tab for each provider (OpenAI, Gemini, Ollama). A dot on each tab and a line on each tile tell you whether that provider is ready: green when it is set up, amber when the provider Prompt Builder uses still needs something, grey otherwise. The Voice input, Writing check and Translation retries cards sit below it, see Writing, voice and translation. You can also type openai, gemini or ollama in the Settings search box to jump to it.

On Ask AI, the Add API key button next to the model picker opens Settings for you when the provider you chose has no key yet.

Saving your changes

The Providers card does not save as you type. When you change a field, the Settings page shows “You have unsaved changes” and a bar at the bottom with Discard and Save changes. You can also press Ctrl+S (Cmd+S on macOS). A toast says “AI provider settings saved.” when it worked. The Test connection button for Ollama works on what is typed in the box, before you save.

Set up OpenAI

Create an API key at platform.openai.com/api-keys (Get a key, next to the API key label, opens it).
In Settings, AI, Providers, open the OpenAI tab under Connections and paste the key into API key (it starts with sk-). The OpenAI tile changes from No key to Key set.
Check Default model. It is pre-filled with the app’s default OpenAI model. Type another model id if you want a different one.
Click Save changes.

The API key field hides the key, with a button to show it. An empty key counts as not set, and you will see “Set an OpenAI API key in Settings first.” in features that try to use it.

Set up Gemini

Create a key at aistudio.google.com/apikey (Get a key, next to the API key label, opens it).
Open the Gemini tab under Connections and paste it into API key (it starts with AIza). The Gemini tile shows Key set.
Check Default model. It is pre-filled with the app’s default Gemini model. Type another model id if you prefer.
Click Save changes.

Once a Gemini key is saved, the Ask AI page loads the list of Gemini models your key can use, so you can pick from it there.

Set up Ollama

Ollama runs models on your own computer (or on a server you control), so there is no API key and nothing leaves your network. Install Ollama from ollama.com and pull at least one model first (for example with ollama pull <model>).

Open the Ollama tab under Connections and check Server URL. The default is http://localhost:11434, which is right when Ollama runs on this machine. If it runs elsewhere, type that address.
Click Test connection. The line under the address says Connected with the Ollama version and the number of installed models, or the Ollama tile turns to Not reachable. A toast tells you the same. If Ollama answers but has no models, the toast suggests pulling one.
Open Default model. The list comes from the server. Use the refresh button beside it to reload the list after you pull a new model. You can clear the choice with the clear control.
Optionally open Advanced and set Context length and Keep model in memory (explained below). Advanced starts folded, and opens by itself when either one holds something other than the default.
Click Save changes.

Context length

Context length is the number of tokens the model keeps in its window (Ollama calls it num_ctx). Leave it empty to use whatever the model ships with. Bigger values let you paste longer text but need more RAM or VRAM. The field only accepts digits.

Keep model in memory

Keep model in memory is how long Ollama holds the model in RAM after a request (Ollama’s keep_alive). Use values like 5m or 1h, 0 to free the memory right away, or -1 to keep the model loaded. The default is 5m. If you clear the field and save, it goes back to 5m.

If Ollama is not reachable

You will see “Could not reach Ollama at <address>. Is it running?” Start Ollama, check the address, and click Test connection again. Local models can be slow to load and answer on a CPU, so Ollama requests are allowed up to 10 minutes before Runova gives up. OpenAI and Gemini requests are stopped after 3 minutes with no answer.

Choose which provider Prompt Builder uses

At the top of the Providers card, the Prompt Builder writes with tiles (OpenAI, Gemini, Ollama or Default CLI) decide what Generate Prompt uses, and the API provider you pick there is also the one the other AI features use when they do not ask you. It uses the key and default model of that provider from Connections. The default is OpenAI.

If the provider you pick is not ready, a note under the tiles says what is missing, with a button that takes you to it: Add OpenAI key or Add Gemini key opens that tab with the cursor in the key field, and for Ollama the note says whether the server is not answering, has no models, or has no model picked.

The tiles also offer Default CLI. With it, Generate Prompt skips the providers and your default AI CLI writes the prompt instead, so no API key is needed. The other AI features keep using the API provider you picked last. See Prompt Builder.

If that provider has no model set, features show “Set a <provider> model in Settings first.” Ollama has no pre-filled model, so pick one in Default model before using it.

Ask AI is the exception. It has its own provider tabs and model picker, so you can switch provider there without touching this setting. See Ask AI.

Which features use these providers

Feature	What it uses
Ask AI (page and popup)	The provider tab and model you pick in the chat. The first time, the model fields start from your Settings defaults.
Prompt Builder Generate Prompt (page and the Build Prompt dialog)	The Prompt Builder provider and its default model, or your default AI CLI when the Default CLI tile is picked.
Project Blueprint (see Projects)	The Written by choice AI provider (Settings), which is the fallback when no suitable AI CLI is installed.
Deploy Assistant	The AI choice AI provider (Settings) in its composer (selected by default). See Deploy: Assistant and logs.
The Ask AI dialog on an SSH terminal and on a Remote Desktop session	The choice AI provider (Settings). Remote Desktop needs a model that can see images, because it sends screenshots. See Remote and Remote Desktop.
The Help page’s Ask the guide chat	Whichever provider you set up here. It also uses that provider’s embedding model to find the right articles, which you can change in Settings > AI > Help search. See Help center.

Translation, voice input and the writing check do not use these providers. They have their own settings, see Writing, voice and translation.

How this differs from AI CLIs

AI CLIs (Claude Code, Codex, Gemini CLI and the others) are separate programs that Runova launches in a terminal. They read and edit your files, run commands and use their own login or subscription. You manage them in AI CLI Manager.

AI providers are plain model calls with an API key or a local server. They answer a question or write some text and do nothing else: no file access, no commands. They are billed to your API account (OpenAI, Gemini) or run free on your hardware (Ollama).

Some features use a CLI and never touch these providers:

Commit message suggestions use the CLI you choose in Settings, Agents, Commit messages.
The Recommended run panel in Prompt Builder asks your default AI CLI (Settings, Agents, Default CLI, or the first installed one) to size a prompt.
Agents you open in the Workspace are CLIs.

If a feature says it needs a CLI, install and sign in to one in AI CLI Manager. If it says it needs a provider, set one up on this page.

Privacy and proxies

Your keys are saved in Runova’s settings on this computer. OpenAI and Gemini requests go straight from Runova to those services, so the text you send is subject to their terms. Ollama requests go only to the server address you entered. If you are behind a proxy, the AI providers follow the proxy settings under Settings, Network.

Tips

Use Ollama if you do not want text to leave your network. Pick a model that fits your RAM, and raise Context length only when you need it.
If Test connection works but chats are slow, the model may be loading. Raise Keep model in memory to 1h or -1 so it stays loaded.
If Ask AI shows a different provider than the one you saved, that is fine. Ask AI remembers its own last choice.