How Cerevisor works

Six things Cerevisor gives you

Any model, any agent tool

Each agent runs on the model that fits the job: Claude Code for the coder, Gemini for research, a local Ollama model for the sensitive step, an API model for the writer. Sign in with a plan you already pay for or bring an API key. Eight provider types, mixed in one workflow.

Every run on the record

Every run writes an append-only record on your machine: 38 kinds of fact, from permission decisions and approvals to tool calls, files touched, web requests and cost. Rerun a single agent, pick up a run that stopped, and answer 'why did the agent do that?' straight from the record.

Visual skill allocation

Skills in ~/.claude/skills and ~/.cerevisor/skills show up in your library. Drag any skill onto any agent on the canvas, or let Cerevisor suggest which skills fit which agent. No config files, no glue code.

Full context control

See exactly what every agent receives before it runs. Inject upstream outputs, drop .md files into input zones, scope folder context per agent, and pre-flight skills. You decide what goes into the context window.

Iterate workflows without rewrites

Loop and Enhance refines a node against its own output, and keeps every version so you can compare rounds and restore one. A run that stopped can be picked up where it left off. A retrospective after each run suggests what to change next.

One view of everything running

Open several worlds at once, two side by side, with their scheduled runs and active agents. See what is running, what is queued, what just finished, and what is waiting for your answer.

Bring your own model

Your keys stay with you. Claude Code, Codex, Cursor and Antigravity run their own steps, so you set their rules before the run. With API and local models, and with Grok Build, Cerevisor approves each action as it happens.

• Anthropic: Claude models through the API. Bring your key.
• Gemini: Google's Gemini models through the API. Bring your key.
• OpenAI-compatible: OpenAI, OpenRouter, Groq, Together, Fireworks, vLLM, LM Studio, and any /v1/chat/completions endpoint.
• Ollama: Local, free, offline. No key required.
• Claude Code: Bundled with Cerevisor. Sign in with your Claude Pro or Max plan, or use an API key.
• Codex: OpenAI's coding agent. Sign in with your ChatGPT plan, or use an API key.
• Cursor: Local Cursor agent on every plan. Cloud Cursor on the paid plans.
• Antigravity: Google's agent tool. Sign in with your Google AI plan.
• Grok Build: xAI's coding agent tool, driven from your own account. Cerevisor approves each of its actions.

System requirements

• Windows: Windows 10 or later, 64-bit. 8 GB RAM recommended. The installer asks for administrator permission. Code signing is in progress, so for now SmartScreen shows a warning on first install.
• macOS: macOS 11 (Big Sur) or later. Apple Silicon and Intel builds available, signed ad hoc. Automatic updates are off on Mac for now; download new versions from the site.
• Linux: AppImage and .deb builds for x64 are published as an early preview.

What stays on your machine

• Cerevisor records usage data on your machine and sends none of it to Cerevisor.
• Your prompts, agent runs, MCP traffic, generated outputs, and provider API keys stay on your machine. Keys are encrypted with your operating system's secure storage.
• Calls to model providers, MCP servers, and web search go directly from your machine to the service you configured.
• Workflows live as plain files (.cerevisor and .cerevisor-world) you can version, share, and audit.
• The app contacts three places on its own: the update check at releases.cerevisor.com (your IP address and a random install id; can be turned off), Lemon Squeezy for a paid license about once a day (license key and a one-way device id, no content), and a one-time 23 MB skill-matching model download from Hugging Face (can be turned off). Full detail is on the security page.

Pricing at a glance

7-day trial of everything on every install, no card required.

Free: €0 (free plan subject to terms; we may change features with 30 days notice) for personal use, students, teachers, and nonprofits. Work for a business, an employer, or paying clients needs Pro or Founding member. Free covers the canvas, the Chat Builder, every supported provider, up to 2 workflows per world and 2 world files.

Pro: €15/mo or €150/yr. Adds MCP servers, the Scheduler, web search, Loop and Enhance, Skill Workshop editing, the Vault, and the ability to trigger Cursor's cloud agents from Cerevisor (your Cursor account, your billing), with up to 5 workflows per world. Requires Cerevisor 3.3.0 or later.

Founding member: €75/mo or €750/yr. Everything with no limits, plus two experimental features: Operator and the team conversation. Anyone who subscribed before Cerevisor 3.3.0 is a Founding member and nothing changes for them.

2 devices per license. 14-day money-back request window on first purchase, processed by Lemon Squeezy.

What you can count on

Real software, shipped today

Cerevisor runs on Windows and macOS today. Everything on this page is in the current release. Features still being shaped are marked Experimental on the pricing page.

Founding member price, locked

Join as a Founding member and your price stays the same for as long as your subscription is active, even if standard prices change later.

Real money-back request window

7-day trial of everything with no card required. First paid purchase eligible for a full refund within 14 days, processed by Lemon Squeezy.

Direct line to the team

Paid members get a direct path to the team for bug reports and feature requests.

If something breaks, we fix it

Your workflows and provider keys stay safe regardless. Tell us what broke and we will fix it.