---
title: "Organizations (hierarchy mode)"
description: "Build a resident team of agents with roles and charters, chat with any of them any time, and let them check with colleagues, remember what matters, and do "
slug: guides/organizations
section: guides
canonical_url: https://cerevisor.com/docs/guides/organizations
last_verified: 2026-07-30
last_verified_version: "1.9.0"
updated_at: 2026-08-14T11:11:55.053396+00:00
---

An **Organization** is a team of agents that lives in your world — not a pipeline you run, but colleagues you talk to. You arrange agents in a top-down chart (like an org chart), give each one a role and a one-line **charter** describing what they own, and then open a chat with **any** of them, any time.

When you ask an agent something, it can:

- **Answer from what it knows** — including things it remembers from your past conversations.
- **Check with colleagues** — it reads the org directory, picks whoever owns the topic, and asks them behind the scenes. You see exactly who it checked with, live.
- **Do real work** — the agent you're talking to has its normal tools, and any **Powers** you've given it (saved workflows) for producing actual deliverables.

Nothing runs while you're away. Agents feel persistent because each keeps a small, durable memory of what matters — distilled after each conversation — but there are no background processes and no idle spend.

## Creating an organization

In the World View, click the **org-chart button** at the bottom of the workflow rail (below the **+**). Pick the folder the org works against, name it, and you land in the org chart editor.

Creating an organization is a paid feature. Opening and chatting with an org someone shared with you is free.

## Building the chart

- **+ Leadership agent** adds a top-level agent (reports to no one).
- **+ Report** on any card adds a direct report under that agent.
- **Drag a card onto another card** to change who it reports to.
- **Tidy** re-lays the chart out top-down.

Every agent needs a **charter** — one line saying what they own (e.g. "Owns competitive research and market briefs"). The charter is what the rest of the org sees when deciding who to check with, so the editor won't save an agent without one.

Each agent also has a role (the same 16 roles as workflow agents), free-form instructions, optional skills, up to 2 Powers, and an optional model provider. Org agents hold live conversations, so they run on chat providers (Anthropic, Gemini, or any OpenAI-compatible endpoint). Providers like Codex or Cursor that run work start-to-finish on their own can still participate — inside the agent's Powers.

## The org bulletin

The pinned card at the bottom-left of the chart is the **bulletin** — context every agent in the org sees (current priorities, conventions, who the customer is). Edit it any time; it applies from the next message onward.

## Chatting with an agent

Click **Chat** on any card. The panel shows the conversation, the running cost against the conversation budget, and — when the agent checks with colleagues — a live list of who it asked and what came back.

A few things to know:

- **Only the agent you're talking to can make changes.** Colleagues it consults can look things up (read files, search the web if allowed) but cannot write files, run commands, or make anything happen. If action is needed, they recommend it and your agent does it.
- **Consultations are capped** (per message, and two levels deep at most), and every conversation has a **budget** — when it's reached, the chat pauses with a clear message instead of quietly spending more. Adjust both in the org's settings.
- **Approvals work like everywhere else.** The agent's tool use goes through the same permission prompts as workflow runs, under your usual approval mode.

## What agents remember

After each conversation, the important parts (facts learned, commitments, work in flight) are distilled into that agent's memory and injected the next time it wakes. Full transcripts stay on your machine for auditing but are never replayed wholesale.

Click **Memory** in the chat panel to see exactly what an agent remembers — and **Forget everything** to clear it. Everything lives locally under your Cerevisor memory folder; nothing leaves your machine.

## Saving and sharing

Organizations save inside your `.cerevisor-world` file alongside your workflows. Use **Export** in the org editor to save a standalone `.cerevisor-org` file you can share; opening one adds the org to your current world.

## Organizations vs. workflows

| | Workflow | Organization |
|---|---|---|
| Shape | Left-to-right pipeline with a Run button | Top-down team you chat with |
| Runs when | You (or a schedule) starts a run | Only while you're chatting |
| Output | Deliverables from a run | Answers + work done during the conversation |
| Memory | Run history and analytics | Per-agent durable memory |

Organizations sit *above* workflows: give an org agent a Power and it runs a whole workflow for you as part of answering.