Agents

An agent is a worker you configure once and use everywhere. Each one has its own personality, its own files, its own tools and memory, and its own settings for which AI engine powers it. You chat with agents, hand them tasks, put them on a schedule, and let them collaborate — and you decide exactly what each one can see and do.

What's in an agent

Every agent is self-contained:

• A persona — the instructions that shape how it works, written in plain language.
• Knowledge — reference documents you give it, kept on hand so it can consult them whenever a job calls for them.
• Tools — the abilities it's allowed to use (files, web, integrations, and more).
• Memory — long-term notes it keeps across conversations.
• Your own context — a private area where you (or the agent, on your behalf) can keep notes and documents for it. They load automatically into your conversations with the agent, separate from its shared knowledge.
• A workspace — the files the agent works in. Depending on its mode, that's a shared team workspace, a private space for each person, or both (see Personal vs. shared below).
• Settings — its name and description, which AI engines and models it uses (and their defaults), its visibility and workspace mode, its department and place in it, which agents it can delegate to or meet with, and which tools it can use. A monitoring view rolls up everything it's doing — scheduled tasks, run history, meetings, notifications, and triggers.

On disk

Under the hood, an agent is a folder, and the pieces above map to real paths you can browse from the agent's file view:

• config/agent.md — the persona. Loaded first into every session.
• config/context/ — markdown/text files that load into every conversation with the agent (1 MB per file, 5 MB total) — operational rules, business context, vocabulary.
• config/user-setup.md — optional per-user onboarding: each new user gets a personal copy that loads only into their chats until they complete it.
• knowledge/ — the reference library: any file type, read on demand, never auto-loaded. Knowledge folders can also be shared between agents as libraries — see Knowledge.
• workspace/ — the shared workspace, in modes that have one.
• users/<name>/workspace/ and users/<name>/context/ — each person's private workspace and personal context. The context files auto-load into that person's conversations only.

The config/ folder is visible to managers only. Everything else follows the agent's mode and roles.

Your first agent

A new install comes with the Personal Assistant — a ready-to-use general assistant, installed automatically the first time you set up. It's there so you can start chatting right away while you set up the rest, and it walks each new user through a short personal onboarding in their first chat.

It's the only agent that ships with the platform. You add the others yourself.

Adding more agents

You build the team you need, three ways:

• Install from the community catalog. Browse ready-made agents from the Agents page and install one in a click. The catalog grows over time.
• Create your own. Start from scratch, give it a persona and knowledge, pick its tools and engine, and you have a specialist tailored to your work — the Creating an Agent guide walks through it.
• Ask an agent to build one. Describe the agent you need in chat and an existing agent interviews you, writes the persona, picks the tools, and installs the new agent with you as its manager (platform creators and admins).

For example, a business might stand up agents for every role — a personal assistant for each person, a support agent, a content agent, a monitoring agent, coding agents for the engineering team — all on one platform.

Zero-config start

When you create or install an agent, OtoDock automatically picks an AI engine you've connected so it works immediately. You can change the engine, model, and behavior any time in the agent's settings.

Who can do what: per-agent roles

Access is granted per agent, so a person can be a power user of one agent and a read-only observer of another. Each assignment carries one of four roles:

Role	What they can do
Viewer	Chat with the agent and work in their own personal space. Read-only on the agent's shared files.
Contributor	Everything a viewer can do, plus add and edit files in the agent's shared workspace. It's the role for an outside collaborator on a project, so a contributor never acts as the agent: no shared tasks, triggers, delegation, shared apps or shared memory.
Editor	Everything a contributor can do, plus act as the agent: shared tasks and triggers, delegation, shared apps and shared memory.
Manager	Full control — configure the agent's persona, tools, knowledge, and settings.

These are separate from platform roles (admin, creator, member), which govern the install as a whole. An admin runs the platform, and a manager runs a specific agent. See Users & Access for the full picture.

Personal vs. shared

Agents can work privately for one person or collaboratively for a group. You pick the mode per agent:

• Personal + shared — each person gets their own private space and a shared workspace the whole team contributes to. The agent works in each person's private space by default, reaching for the shared one when needed. The most flexible mode, and a good default for most collaborative agents.
• Shared + personal — the same two spaces, but the agent makes the shared workspace its home base by default, while each person still has a private space it can use.
• Personal only — every person gets a completely private agent. Conversations, files, and memory are theirs alone and never shared. Ideal for a personal assistant.
• Shared only — one shared space and one shared history for everyone. There's no private per-person area. This fits agents that act on behalf of the whole organization, like a support agent the whole team leans on. Every chat here runs as the agent itself, so only editors and managers can chat there. Viewers and contributors can read the chats, but they can't write in them. To let them chat, pick a mode with personal chats.

Switching modes never deletes anyone's files — it only changes what's shared going forward.

Working together

Agents don't have to work alone. Three features let them combine forces:

• Delegation. An agent can hand part of a job to another agent that's better suited to it, then fold the result back into its own work. You choose which agents each one is allowed to delegate to, or let departments wire it up for you.
• Meetings. Bring several agents into one conversation to tackle a problem from different angles, with a facilitator keeping things on track.
• Departments. Organize agents into an org structure with levels. Each department decides which of its agents can hand work to which, and the Agents page shows the whole company on a live map.

→ See Departments, Meetings, and Tasks for the details.

Next steps

• AI Engines → — choose what powers each agent.
• Departments → — turn a roster of agents into an org chart.
• Using Tools (MCPs) → — give agents real abilities.
• Memory → — how agents remember across conversations.
• Users & Access → — platform roles, SSO, and managing people.