---
title: "About Cerevisor - Local-first control plane for AI agents"
description: "Cerevisor is a local-first harness for agentic work. We help operators, founders, and teams turn scattered AI prompts, tools, and agents into workflows they can run, audit, and trust. Not a consultancy, a product."
canonical_url: https://cerevisor.com/about
type: page
---

# We're building the control layer for agentic work.

Cerevisor helps teams turn scattered AI prompts, tools, agents, and scripts into workflows they can run, inspect, improve, and trust.

## Why Cerevisor?

Cere- from cerebrum: cognition, strategy, the part that plans. -visor for vision, a lens that filters the noise, and the instinct to supervise the work rather than hand it off blind.

Not another chatbot. Not a consultancy. Cerevisor is a local-first control plane for AI agents: the place where the prompts, tools, agents, and scripts your team already runs become workflows you can run, inspect, improve, and trust. Your keys and your data stay on the machine.

## Built by Operators

We run multi-agent workflows every day. The harness is shaped by what actually holds up in real work, not by what looks good in a demo.

- We run multi-agent workflows daily, so the harness is built around real work, not demo-ware
- We stay deep in the model ecosystem: new providers, new harnesses, new MCP servers, the week they ship
- We know what agents can and cannot be trusted with yet, and we build the guardrails for the gap
- Local-first is a principle, not a setting: your prompts, keys, and runs stay on your machine
- We judge the product by one question: can your team rerun the work next week and trust the result?

## Mission

Make agentic AI something a team can actually govern: humans in the loop where judgment matters, prompts and keys on the machine, and every run replayable and trusted.

## Vision

Every AI-enabled team runs like a well-instrumented one: clear inputs, auditable outputs, continuous learning loops, and people whose expertise is amplified by AI, not replaced by it.

## Design Principles

The values baked into the Cerevisor harness

- **Local-First by Default**: Prompts, runs, MCP traffic, and provider keys stay on your machine. Nothing flows back to Cerevisor.
- **Provider-Neutral**: Anthropic, Ollama, OpenAI-compatible endpoints, and Codex CLI all work. Mix providers per agent in a single workflow.
- **Audit-Ready by Design**: Every run lands as an NDJSON trail you can replay, diff against last week, and answer 'why did the agent do that?' from data.
- **Human-in-the-Loop**: Approval checkpoints, escalation paths, and review steps belong on the canvas wherever judgment matters.
- **Composable Files**: Workflows and worlds are plain files (.cerevisor and .cerevisor-world) you own, version, share, and re-run.
- **Visual, Not Limiting**: A canvas-first experience for operators who do not write code, that stays out of the way of those who do.

## What the Harness Gives You

Why operators pick it over duct-taping their own agent stack

- Runs locally on your machine, your provider keys never leave
- Multi-model orchestration: mix Anthropic, Ollama, OpenAI-compatible, and Codex CLI per agent
- Full NDJSON audit trail for every run, replayable and diffable
- MCP-aware runtime out of the box, no glue code to maintain
- Plain-file workflows and worlds you can version with git
- Free to start with a 7-day full Pro trial, no card required