Models & LLM connections

Every agent runs on an AI model. RAiN AI Studio follows a simple policy: Claude models are included — they run on RAiN's own account with zero setup — and other major providers work with your own key.

The default: Claude, no setup

New agents start on Fast (Haiku) — quick and cost-efficient, ideal while you're designing and testing. In the agent's editor you can switch to Balanced (Sonnet) or Most capable (Opus) for harder work. All three run on RAiN's key: nothing to configure, nothing to pay a provider.

Using other providers (your own key)

Prefer GPT, Gemini, Grok, or an OpenAI-compatible service (Groq, DeepSeek, Together, Ollama, …)? Connect your own API key once and every agent in your workspace can use that provider.

The quick way — right where you pick the model:

Edit an agent and open the Model dropdown. Other providers' models are listed; ones without a key say (key needed).
Pick one — an API key panel appears in the form.
Paste your key (OpenAI-compatible services also ask for their base URL) and save the agent.
Done — every agent in this project can now use that provider.

The management view — LLM Connections (sidebar): see every provider, which are connected, replace a key, or Disconnect one.

Designing with your own model

Connect a provider and it doesn't only run your agents — it designs them too. Build with AI uses your own model when this project has one connected, and falls back to RAiN's Claude when it doesn't. Nothing to switch on.

GPT connected → designs on GPT.
Gemini connected → designs on Gemini.
Nothing connected → designs on RAiN's Claude, included as always.

One exception: an OpenAI-compatible endpoint is used for running agents but not for designing. RAiN can't know which model name your endpoint serves, and guessing one would fail at your provider rather than here — so designing stays on Claude and your runs stay on yours.

How your keys are protected

Keys are encrypted at rest and only ever decrypted at the moment your agent runs or a design is drafted.
A stored key is never shown again — not to you, not to anyone. Replacing it means pasting a new one.
Keys are per project, and never shared with another expert. A key you add in one project stays there: connect the provider again in each project that needs it. That's deliberate rather than an oversight — a project you hand over to a client shouldn't be able to send from your mailbox or spend on your key.

Who pays for what

Model	Runs on	Cost
Claude (Haiku / Sonnet / Opus)	RAiN's key	Included with RAiN AI Studio
GPT, Gemini, Grok, compatible services	Your key	Billed by that provider to you
Designing with Build with AI	Your key if this project has one, otherwise RAiN's Claude	Included unless it's your key

FAQs

Which model should I use? Design and test on Fast (Haiku). Move up to Sonnet/Opus (or a provider you prefer) when a task needs deeper reasoning — you can set a different model per agent.

My run failed with "No … key connected". The agent uses a provider you haven't connected (or the key was removed). Edit the agent and either add the key or switch to a Claude model.

Is my key safe? It's encrypted before it's stored, never returned by any screen or API, and only used to run your agents. You can disconnect it at any time under LLM Connections.

What happens if I disconnect a provider? Agents still set to its models will fail their next run with a clear message until you reconnect a key or switch them to Claude.

Can I use one key for the whole team? No — keys are per project, and each expert connects their own. Add the provider again in each project that needs it.

Why did Build with AI use Claude when I've connected GPT? The key is connected in a different project. Designing reads the credentials of the project you're designing in, so connect the provider here too and the next design will use it.