In this section

Connect the AI assistant

The assistant needs an AI model to think with. Connect one of two ways: use an AI app already on your computer, or paste an API key. Works with Claude, Gemini, Codex, and more.

Which one?

You haveDo this
Claude Pro or MaxUse Claude Code.
ChatGPT Plus or ProUse Codex CLI.
A Google accountUse Antigravity for Gemini.
An API keyPaste it.
Ollama on your computerPick Ollama. No key needed.
Nothing yetGet an API key from a provider, or install Claude Code.

Option 1: use an AI app on your computer

Tovu runs a coding-agent app installed on your computer. It uses that app's own sign-in and plan. Tovu never sees your password. In the admin this is called Local CLI.

With Claude Code:

  1. Install Claude Code. On a Mac, open Terminal and run curl -fsSL https://claude.ai/install.sh | bash.
  2. Sign in. Run claude and sign in with your Claude account.
  3. Open the setting. Admin → Settings → Execution mode → Local CLI.
  4. Find it. Click Rescan. Claude Code shows up with its version number.
  5. Pick it. Choose a model, or leave Default.

Other apps work the same way: Codex CLI (sign in with ChatGPT), Antigravity (sign in with Google; runs Gemini), OpenCode, Cursor Agent, GitHub Copilot CLI, Qwen Code, Kimi CLI, Aider, and more. Tovu lists every one it finds. Sign in to each one in a terminal first.

The app must be on the computer that runs Tovu. With the desktop app, that's your computer. On your own server, install and sign in on the server.

Why there's no "Sign in with Claude"

Anthropic doesn't let other apps sign in with a claude.ai account, so Tovu can't offer it. Claude Code is how your Pro or Max plan reaches Tovu.

A Claude plan doesn't include API credits. An Anthropic API key is billed separately.

Option 2: paste an API key

Tovu calls the provider directly with your key. In the admin this is called BYOK (bring your own key).

Supported: Anthropic, OpenAI, Azure OpenAI, Google Gemini, OpenRouter, and Ollama.

  1. Get a key from your provider: Anthropic Console, OpenAI, Google AI Studio, OpenRouter, or the Azure portal.
  2. Open the setting. Admin → Settings → Execution mode → BYOK.
  3. Pick your provider.
  4. Paste the key and click Save key.
  5. Pick a model, click Test connection, then Save settings.

Azure: also enter your resource's endpoint as the Base URL. Ollama: no key. Tovu looks for it at http://localhost:11434/v1.

Who bills you

OptionBilled by
Claude CodeYour Claude plan. Its usage limits apply.
Other appsThat app's own account or plan.
API keyThe provider, per use.
OllamaNobody. It runs on your computer.

Tovu charges nothing for the assistant.

The chat on your public site uses its own key

The key above powers the assistant in your admin only. The chat visitors see on your site needs its own key: Admin → AI Assistant → Visitor's AI Assistant. Every visitor conversation is billed to that key.

Keep your keys safe

  • Keys are stored encrypted on this Tovu install. The admin shows only the last 4 characters.
  • Publishing never copies your keys to your live site.
  • Set a monthly spending limit in your provider's console.
  • Paste keys only into the key field. Never paste one into the assistant chat.
  • Key leaked? Delete it in your provider's console, make a new one, and paste that in.

Troubleshooting

"No usable BYOK credential." You're in BYOK mode with no saved key or no model. Save a key and pick a model.

The app says "Not installed." Install it, sign in once in a terminal, then click Rescan. On your own server, install it on the server.

The app is found but replies fail. Its sign-in may have expired. Run it in a terminal and sign in again.

The key is rejected. Check you picked the right provider. Tovu warns when a key looks like another provider's. New API accounts often need billing set up before keys work.

"The server cannot store keys yet." Your server has no encryption key. Set TOVU_INTEGRATIONS_ROOT_KEY and restart. This only happens on your own server.

Local CLI stopped answering. Admin → AI Assistant → Admin AI Assistant → Restart assistant.

Next: what the assistant can change, and how it asks first.