AI Providers and API Keys

Caiioo connects to 500+ AI models through a small set of providers. You bring your own API key — that means you pay the provider directly, and Caiioo never touches your spending.

Supported Providers

Five providers are generally available today:

OpenRouter (default) — a single key unlocks 500+ models including Anthropic Claude, Google Gemini, OpenAI GPT, Mistral, Groq, Llama, and more. This is the recommended path for most users — Caiioo's tools expect OpenRouter's model registry, and you only manage one key.
Google AI — a direct connection to the Gemini API with your own Google AI key. Google offers a genuine free quota, so this is a real no-credit-card way to run a frontier model — with native thinking and up to 2M context. Settings walks you through getting a free key.
xAI (Grok) — direct access to Grok models via api.x.ai. Connect with an xAI console API key, or simply sign in with your Grok account (SuperGrok or X Premium+ subscription) — approve access in the browser and you're connected, no key to copy. Note: xAI enables API access per account and subscription tier on their side; if chats fail with a permission error after connecting, use an xAI API key instead.
CometAPI — 500+ models (GPT, Claude, Gemini, DeepSeek, and more) through one pay-as-you-go CometAPI key. The model picker shows benchmark-tier medals for its catalog, and total CometAPI usage is tracked in Settings.
Ollama — runs open-source models locally on your machine. Fully private, no internet required, free. Useful for offline work or when you want absolute control over inference. (Ollama Cloud also works with an optional API key.)

Direct connections to other providers (Anthropic, OpenAI, Poe, MLX, and more) are in testing and not yet generally available. For now, use those models through your OpenRouter key — that's what OpenRouter is for.

See Terms of Use for policies.

Get Started (Two Options)

Option 1: Use the Free Provisioned Key

We provide a free OpenRouter key that lets you try Caiioo immediately. It has access to select free models with usage limits — perfect for testing things out.

To keep using it: just open Caiioo and start chatting. No setup needed. Pro subscribers get an additional $5/month OpenRouter credit bonus auto-applied to this same key, on top of whatever balance is already there.

While you're on the included key, the OpenRouter key field in Settings says so instead of looking empty — you'll see a note that you're using Caiioo's account key and your remaining balance. The included key stays on the device you set it up on; your own OpenRouter key is also stored per-device and only appears on other devices after you turn on Private Sync.

Credit-consuming tools—image, video, and music generation, advanced OCR, and Perplexity AI search—need OpenRouter credits, not just a key. On the free provisioned key these tools report "no credits." You have two ways forward: enter your own OpenRouter API key (BYOK) in Settings > AI Providers, or upgrade your Caiioo plan—a paid plan tops up this same provisioned key with credits each month, so these tools just work.

Option 2: Bring Your Own Key (BYOK)

Want full access without the free-tier limits? Add your own key.

OpenRouter (recommended):

Sign up at openrouter.ai and create an API key
Go to Settings → AI Provider
Select OpenRouter and paste your key — or click Sign in with OpenRouter to connect your account without copying anything
That's it — the key stays on your device and never reaches Caiioo's servers

Google AI (direct Gemini):

Go to Settings → AI Provider and select Google AI
Follow the guided "get your free key" steps (Google AI Studio issues keys free, no credit card)
Paste the key — it stays on your device and requests go straight to Google

Ollama (local):

Install Ollama from ollama.com and pull a model (ollama pull llama3.2)
Go to Settings → AI Provider
Select Ollama and confirm the server URL (default http://localhost:11434)
Pick any model you've installed locally

When you pick a local chat model, AI-powered tools that would send your data to a remote AI provider prompt for your approval first. See Privacy & Data → Remote AI Providers.

Model Configuration showing Default Model and Max Steps

Quick Tasks Model

A separate, cheaper model that Caiioo uses automatically in the background for things like generating thread titles, classifying intent, picking which tool to call, and other short helper queries — you never prompt it directly. Configure it in Settings → Model Config (a small OpenRouter free model or a local Ollama model is plenty). If not set, the system falls back to your default model.

BYOA (Bring Your Own Auth) for Tools — Pro

BYOA is different from BYOK. It's not for LLMs — it's for tool connections (Google Workspace, Microsoft 365, GitHub, etc.). At Pro, Settings → Account → Account Authorizations and Permissions lets you plug in your own OAuth client when your organization won't whitelist Caiioo's default client ID:

Use Your Own Google OAuth App — walk through creating a Google Cloud project and pasting in the client ID + secret
Use Your Own Microsoft Azure App — same flow for Azure-registered apps
Add Custom OAuth Provider — generic OAuth 2.0 flow for any provider that supports it

The wizards write the credentials straight to the Credentials Vault. They never leave your device.

API Key Security

Your API keys are stored locally on your device. They're never sent to Caiioo servers. When you make a request, your key goes directly to the AI provider. See Privacy & Data for the full story.

Performance Depends on Your Model

Caiioo handles orchestration — tools, context, multi-step workflows. But the quality of the AI's responses comes down to which model you choose.

For complex tasks (research, analysis, long documents): use a frontier model — Claude Sonnet, GPT-class, or Gemini Pro — all accessible through your OpenRouter key. They handle multi-step reasoning and tool use best.

For quick answers (questions, summaries, simple edits): fast models — Claude Haiku, smaller GPT or Gemini Flash variants — are cheaper and faster. Perfect for everyday tasks.

For experimenting: OpenRouter lets you try 500+ models with one API key. Find what works best for your workflow without committing to a single provider.

A note on the Creativity slider: the newest models (like Claude Sonnet 5 and later) manage their own creativity and don't accept a temperature setting. When you select one, the Creativity slider in Model Configuration is disabled with a short note — that's expected, not a bug.

For offline / fully-private work: Ollama runs open-source models on your machine. Slower than cloud frontier models, but nothing leaves your hardware.

Caiioo's tools, context layers, and orchestration amplify whatever model you pick. A good model becomes great when it has the right context and tools to work with.