Connect a model provider

Cerevisor doesn't ship its own model. You connect it to a provider you already have access to. Several kinds are supported, and you can have any number of each (except the singletons noted below).

On a fresh installation, choose an example or describe your task in the first-project guide, then select Connect my AI. You can also open provider setup any time from the title bar menu or via Settings → Providers → + Add provider. Closing setup during the guide returns you to your project without losing the brief.

This page walks through the simplest setup for each provider type. For the deeper guide, including managing multiple providers, per-agent overrides, and library administration, see Guides → Providers.

Start with what you already have

The setup pane opens by asking one question — What do you already have? — with five answers:

Answer	Where it takes you
ChatGPT Plus or Pro	The ChatGPT Plus / Pro tile (small label: via Codex). Uses the ChatGPT plan you already pay for.
Claude Pro or Max	The Claude Pro / Max tile (via Claude Code). Uses your Claude plan, or an Anthropic key.
A Google account	The Google AI plan tile (via Antigravity). Gemini, Claude and open models on your Google plan.
An API key	Scrolls the key column into view — Anthropic, OpenAI, Gemini, xAI, OpenRouter, Cursor, and anything else that speaks the same API.
Nothing yet — run free on this computer	The Ollama tile. Free models that run on your own machine.

Every tile says what you need and roughly how long it takes before you commit to it, and shows a live status chip — Ready — already signed in, Ready, Quick install needed, or Needs attention. The tiles the chooser can point at are shown first; the rest wait behind More options, and Fewer options puts them back.

Whatever you type stays on this machine. Cerevisor never sends your sign-in or your keys anywhere else.

What happens when you connect

The same three things happen in every lane, so you always know where you are:

• Installing is one click where it is possible. For ChatGPT Plus / Pro and for the Google AI plan, Cerevisor runs the install for you, shows exactly what it runs, and tells you plainly how it went — including when the install worked but the tool still is not findable yet. Nothing here needs admin rights, and you are never asked to open a terminal.
• Signing in opens your browser, and the wait can end. After a minute Cerevisor offers help without giving up; after five minutes it stops waiting and says so rather than spinning forever. Cancel ends the wait at any point and puts the Sign-in button back.
• Save waits for a passed connection check. While a check is still needed the Save button is greyed and says why in one line — Install it first., Sign in first., Checking your connection…, Waiting for the connection check. If the check fails, the button becomes Save anyway so you can still keep the connection and sort the problem out later. Editing a connection you already saved is unchanged — that Save is always available.

A connection check passes only on proof: the tool exited cleanly and gave a real answer. A test that times out, or comes back signed-out, or returns nothing, is reported as a failure rather than quietly passing.

When the first connection saves, a green banner reads You're connected and names what will run your workflows, with Build my first workflow beside it and Add another if you want a second connection first.

Choosing models afterwards

The setup pane's left side keeps one control while you have a single connection: Default model — the model new workflows use. Everything else (a separate model for background work, a separate model for the chat builder, what to fall back to when a plan hits its limit, a daily spending cap, reasoning effort) sits under More model settings, which appears once you have a second connection to choose between — or straight away if any of those settings is already set, so a value can always be seen and changed where you set it.

The provider types

Type	What it is	When to use it
Anthropic	Direct API connection to Claude (Opus, Sonnet, Haiku). Cerevisor runs the agents itself and asks you before anything risky.	You have an Anthropic API key and want the most capable hosted models with full Cerevisor control over every step. Singleton: one entry.
Gemini	Direct API connection to Google's Gemini (2.5 / 3.x families). Cerevisor runs the agents itself.	You have a Google AI Studio API key and want Gemini's long-context Pro models or its cheap Flash workhorses. Singleton. See the Gemini provider page for the dedicated walkthrough.
Claude Agent SDK	Anthropic's Claude Agent SDK, running inside Cerevisor. Claude Code runs the work in its own engine while Cerevisor supervises.	You want Claude Code-style agent execution (read/write/edit/bash/glob/grep tools out of the box) without leaving Cerevisor. Works with an API key or a Claude Pro/Max subscription. Singleton. See the Claude Agent SDK page.
OpenAI-compatible	Any endpoint that speaks the OpenAI Chat Completions API. Includes Ollama, OpenRouter, xAI, Together, Groq, Fireworks, vLLM, LM Studio, and OpenAI itself.	You want local models (Ollama), routing across providers (OpenRouter), or anything OpenAI-API-shaped. Multiple entries allowed.
OpenAI Codex CLI	OpenAI's Codex command-line tool. Codex runs the work in its own engine while Cerevisor supervises.	You have a ChatGPT subscription (Plus/Pro) and want to use Codex's quota rather than paying per token. Singleton.
Cursor Agent	Cursor's agent. Cursor runs the work in its own engine while Cerevisor supervises.	You have a Cursor account and want Cursor's agent (with its tool catalog and optional cloud execution). Singleton.
Antigravity	Google's Antigravity command-line tool. Antigravity runs the work in its own engine while Cerevisor supervises.	You have a Google AI plan and want its quota to power the agents. Antigravity is multi-model: it runs Gemini, Claude, and an open-weight model on your Google plan, with no API key. Singleton.

Setting up Anthropic

You'll need an API key from console.anthropic.com.

1. Answer An API key, then click the Anthropic tile.
2. Paste your API key.
3. Click Test connection. Cerevisor asks Anthropic for your model list to confirm the key works, and fills the model picker from the answer.
4. Click Save.

The key is stored in your operating system's keychain, never in plain text on disk.

Setting up Ollama (local models)

You'll need Ollama installed and at least one model pulled (ollama pull llama3.1, for example).

1. Answer Nothing yet — run free on this computer, or click the Ollama tile.
2. Base URL: http://localhost:11434/v1
3. API key: leave blank, or enter any non-empty string: Ollama doesn't check.
4. Click Test connection.
5. Pick a default model from the dropdown.
6. Click Save.

You can repeat this with different base URLs to add OpenRouter, Groq, Together, or any other OpenAI-shaped endpoint as separate library entries.

Setting up ChatGPT Plus / Pro (via Codex)

You'll need a ChatGPT Plus or Pro plan. There is nothing to install by hand — Cerevisor installs the small helper for you. Takes about two minutes.

The screen is two steps, with everything that already has a sensible default folded under Advanced.

1. Answer ChatGPT Plus or Pro, or click the ChatGPT Plus / Pro tile.
2. Step 1 — Install Codex. If it isn't on this computer yet, click Install Codex: Cerevisor runs the install for you and shows you everything it does. Codex needs Node.js on this computer; if that's missing, Cerevisor tells you. When it finishes you'll see Codex is installed. Next: sign in. — or, if the install worked but the helper still can't be found, Installed, but Cerevisor can't find it yet. Click "Check again".
3. Step 2 — Sign in. Click Sign in with ChatGPT. Your browser opens; approve with your ChatGPT account. If the browser doesn't work for you, choose Use a code instead for a link and a code you can enter by hand. Cerevisor notices when sign-in completes — and also when it stops early, so you're never left waiting on something that already ended.
4. Click Save & finish once the connection check has passed. If the check fails, the button reads Save anyway.

Advanced holds what you normally don't need to touch: execution permissions, the Windows sandbox setting, storing the sign-in in your operating system's keychain, the default model, the connection check, and the full install log. The defaults are the same whether or not you open it.

Cerevisor never stores your OpenAI sign-in — Codex owns its own.

Windows note: an account folder with a space in it (C:\Users\Test User) used to read as signed out forever; it doesn't any more. If Codex is installed after Cerevisor was launched, Cerevisor now finds it without a restart on Windows. On macOS, if you install Node.js after launching Cerevisor, restart Cerevisor.

Setting up Gemini

You'll need an API key from aistudio.google.com/apikey.

1. Answer An API key, then click the Gemini tile.
2. Paste your API key.
3. Click Test connection. Cerevisor sends one tiny request and checks that a real answer comes back — an empty reply counts as a failure, not a pass.
4. Click Save.

For the full walkthrough (model list, pricing quirks, thinking budgets), see the Gemini provider page.

Setting up Claude Pro / Max (via Claude Code)

Use your Claude Pro or Max plan, or an Anthropic API key from console.anthropic.com. There is nothing to install: Cerevisor carries its own copy of Claude Code, and uses yours instead when yours is newer. Takes about a minute.

1. Answer Claude Pro or Max, or click the Claude Pro / Max tile.
2. Pick how you want to sign in: Claude plan (uses your Pro/Max plan) or Anthropic API key (billed per token).
3. On a Claude plan, click Sign in with Claude. Your browser opens; Cerevisor captures the sign-in for you — no terminal, nothing to paste. If you're already signed in to Claude on this computer, most of this happens without you doing anything. The wait offers help after a minute, stops after five, and Cancel ends it at any time.
4. With an API key, paste your key. (It lives in its own keychain slot, separate from the Anthropic connection's, so you can use the same key for both without conflict.)
5. Click Test connection. With an API key this runs one real question-and-answer, because a key that works doesn't prove Claude Code can start on this computer — so it can take up to two minutes the first time, and it says so while it waits. If the key works but Claude Code can't start here, that's what the message says.
6. Click Save.

For the current limitations (no per-tool prompts, no skill injection, etc.), see the Claude Agent SDK page.

Setting up Cursor Agent

You'll need a Cursor account with an API key generated at cursor.com/dashboard.

1. Answer An API key, then click the Cursor tile.
2. Paste your Cursor API key.
3. Click Validate. Cerevisor asks Cursor to confirm the key.
4. Run the connection check. Cerevisor asks Cursor for one trivial answer; it counts as a pass only when Cursor reports the work finished and sends real text back.
5. Click Save.

If Cerevisor tells you this copy of the app is missing the part that talks to Cursor, reinstall Cerevisor from cerevisor.com — your workflows and settings are kept. You are never asked to repair an installed app yourself.

Cursor runs can also execute in the cloud, see Guides → Advanced → Cursor cloud runs for the full setup.

Setting up a Google AI plan (via Antigravity)

You'll need a Google account with an AI plan. There is nothing to install by hand. Takes about two minutes.

1. Answer A Google account, or click the Google AI plan tile.
2. If the helper isn't on this computer yet, click Install — Cerevisor runs Google's own install for you, shows exactly what it runs, and streams the output. It installs to your user folder, so no admin rights are needed. (Prefer your own terminal? The panel shows the command with a Copy button, and links to antigravity.google.)
3. Click Sign in. Your browser opens to sign in with your Google account. The wait offers help after a minute, stops after five, and Cancel ends it at any time.
4. The panel finishes by itself. When the sign-in comes back, Cerevisor runs the connection check on its own — you don't press anything. Then pick a default model: the list includes Gemini Flash/Pro, Claude Sonnet/Opus and an open model, each at a chosen effort level.
5. Click Save.

Cerevisor never stores your Google tokens; Antigravity owns its own sign-in. One thing to know up front: Antigravity's output carries no usage counts, so Cerevisor can't show a cost figure for these runs — Analytics stays blank for them, which is expected.

Setting a default

When you have more than one provider in the library, one of them is marked as the default. New workflows use the default automatically; existing workflows keep whatever provider they were saved with.

To change the default: Settings → Providers, then click Set default on the entry you want.

Removing a provider

Settings → Providers, then click the trash icon on the entry you want to delete.

If any open workflow has an agent that references the credential you're trying to delete, Cerevisor blocks the deletion and shows you which agents would break. Either change those agents to a different provider first, or close the workflow.

Where to go next

→ Build and run your first workflow