---
url: 'https://docs.mychatbot.app/agents.md'
description: >-
  The routine-first mental model for the Agents Platform: routines orchestrate
  repeatable work while agents, skills, connectors, knowledge, and models
  execute it.
---

# Agents: concepts

The Agents Platform is **routine first**. A routine owns repeatable or autonomous work;
agents and models execute its leaf steps, while skills, connectors, and knowledge give
those steps the capabilities they need. Conversations and the lobby remain useful for
ad-hoc work, prototyping, inspection, approvals, and debugging.

Your workspace also ships with a curated set of **built-in agents**, each tuned for part
of the business. You can customize any of them or build a blank **custom agent**. Read the
[operating model](/agents/operating-model) first when designing a production process.

::: tip In the app
Everything here lives under **Agents** in the left nav → <https://app.mychatbot.app/agents>. Use the [lobby](/agents/lobby) (`/agents/chats`) for conversation work and the [Tasks page](https://app.mychatbot.app/agents/tasks) for routines and automation. The [roster](/agents/library) (`/agents`) lists every agent available to execute work.
:::

![Your agents roster — your custom agents, your AI team, and the agent library](/screenshots/agents-list.png)

## Cheat sheet

| Term | What it means |
|---|---|
| **Built-in agent** | A ready-made agent MyChatBot ships, tuned for a job — it comes with a default model, prompt, tools, and skills. You can't rename the built-in *kind*, but you can fully customize your copy of it. |
| **Customization** | The changes you make to an agent. Only the fields you touch change; everything else stays at the built-in default. Scoped to your account and marked with a **Customized** badge. |
| **Reset to default** | One button that clears all your customizations and reverts the agent to how it ships. |
| **Custom agent** | A blank-canvas agent you build from scratch. You get up to **10**. See [Custom agents](/agents/custom-agents). |
| **Chief of Staff** | The [lobby](/agents/lobby) front door — a light triage agent that answers quick questions itself and hands the rest to a specialist, in the same thread. |
| **Coordinating agent** | An agent that runs a team of specialists behind a single chat, instead of doing the work alone (e.g. **Content Director**). |
| **Routine** | The durable orchestration unit for repeatable work. It can sequence, branch, loop, fan out, schedule, or react to supported events, while configured agents execute its leaf steps. See [Routines](/agents/routines). |
| **Trigger** | "When X happens, run Y" — starts a routine or agent from a secret web link or a supported **external connected-app** event. Native Sales Platform changes do not fire Agents triggers today; use a scheduled polling routine for those. See [Triggers](/agents/triggers). |

### Agents that ship in your workspace

Your **AI team** comes with the workspace: the [lobby](/agents/lobby) front door plus a set of general-purpose and specialist agents. A separate **agent library** holds specialists you add only when you need them, and you can build **up to 10 custom agents** of your own.

| Agent | Default model | Built for |
|---|---|---|
| **Chief of Staff** | Gemini Flash | The lobby front door — triages your request and routes it to the right agent |
| **Sales Platform Wizard** | GPT-5 | Sales Platform setup & operation — the main sales-side agent |
| **Personal Assistant** | Claude Opus | General knowledge work — browsing, code, research, and building your own skills and [custom routines](/agents/custom-routines) |
| **Bulk Text Worker** | GPT-5 mini | High-volume, low-cost text — classify, extract, normalize, tag |
| **Content Factory** | Claude Sonnet | Marketing media — product photos, video ads, reels & music, with a full [creative studio of ready-made formats](/agents/creative-studio) |
| **Site Builder** | Claude Sonnet | Single-page sites published to a live `*.page.mychatbot.app` address |
| **Content Director** | Claude Sonnet | Coordinates a Content Team of three creative specialists (text at scale, images & media, live pages) behind one chat |

The **agent library** adds specialists like **HR Recruiter (Work.ua)** and **PROM Cart Designer** once you add them to your team — see [Your team & the agent library](/agents/library) for the full, current roster and how adding/removing works.

::: tip Specialists are just pre-tuned agents
Nothing about the built-in agents is special under the hood — each is an ordinary agent with a tuned model, prompt, tools, and skills. Two library agents (HR Recruiter, PROM Cart Designer) started as one customer's custom agent and were promoted into the shared library. Everything below — customizing, resetting, pausing — works the same on all of them.
:::

## Customizing an agent

Open an agent, then go to the **Settings** tab (<https://app.mychatbot.app/agents> → pick an agent → **Settings**). Anything you change here layers on top of the built-in defaults and applies **only to your account**. Leave a field empty and it falls back to the platform default.

![The Settings tab: Identity (name, avatar, model, labels), What it can do (tools), Connectors, and Skills](/screenshots/agents-tools.png)

| In the app | What it controls | Default when left empty |
|---|---|---|
| **Settings → Identity → Name / Description / Avatar** | How the agent is labeled in your list | The built-in name and look |
| **Settings → Identity → Model** | Which model powers it | The built-in default (shown as "· default") |
| **Settings → Identity → Labels** | Filter pills on your agents list | none |
| **Instructions** tab | What it focuses on, what to avoid, how it responds — layers on top of its built-in expertise | The built-in prompt |
| **Settings → What it can do** | Which tools the agent can reach for (on/off toggles) | The built-in toolset |
| **Settings → Connectors** | Which of your connected apps it can use | The built-in connectors |
| **Settings → Skills** | Custom abilities attached to it | The built-in skills |

How your changes combine:

* **Tool and connector toggles are independent** — turning one off doesn't touch the others. Off means the agent won't see that tool at all; it answers from memory and the conversation only.
* The **Instructions** you write replace the built-in prompt for your copy of the agent.
* The **Skills** and **Labels** lists show exactly what's attached; you edit them directly. If you remove the built-in entries there, they're gone for your agent — so re-add any you still want.

Hit **Save** and a **Customized** badge appears on the agent.

::: tip Your changes take effect on the very next message
Customizations aren't baked in — they're read fresh on every run. Change a setting, save, and the next message already uses it. Your changes never affect other accounts running the same built-in agent.
:::

::: tip You can also let the agent edit itself
If the **Self-edit** tool is on, you can just tell the agent in chat to change its own instructions, model, tools, name, or skills ("be more terse", "turn on web search"). Every self-change pauses for your approval before it sticks.
:::

### The model is a default, not a lock

Every agent ships with a default model (marked "· default" in the picker), but you can move **any** agent to a different model from **Settings → Identity → Model**. Cheaper model for a high-volume job, a frontier model for hard reasoning — it's a per-agent choice. Custom agents also let you pick the model right when you create them.

### Reset to default

The **Reset to default** button on the Settings tab clears every customization and reverts the agent to how it ships. If you haven't customized anything, there's nothing to reset (the button is a no-op — "running on platform defaults").

::: warning Toggles reset one at a time; lists reset as a whole
Turning one tool off leaves your other tools alone. But the **Skills** and **Labels** lists are edited as a single list — if you clear one and add a single entry, you've dropped everything the agent had by default. Include the ones you still want.
:::

## Custom agents (build your own)

When no built-in agent fits, build one from scratch — a blank-canvas agent that's
yours alone, with a name, model, and instructions you choose. You get **up to 10**.
The full flow (creating, deleting vs. resetting, graduating into the library) is on
its own page: **[Custom agents](/agents/custom-agents)**.

## Coordinating agents (teams)

Some agents **coordinate a team of specialists** behind a single chat rather than doing the work alone. The shipped example is **Content Director**, which runs a **Content Team** of three creative specialists: **Bulk Text Worker** (variants, captions, hooks at scale), **Content Factory** (images and other media), and **Site Builder** (landing pages published live).

To you, a coordinating agent behaves like one agent:

* one avatar, one chat thread;
* it routes each request to the right specialist and returns a single synthesized reply;
* replies are attributed to whichever member produced them, so you get "the team" without juggling separate chats.

Tools, skills, and connectors are configured **per member, not on the coordinator**: open any member to flip its toggles, and those changes apply wherever that member is used. The coordinator's own instructions act as its "how to delegate" playbook.

::: tip The lobby is the main conversation surface
Day to day, ad-hoc multi-agent work flows through the [lobby](/agents/lobby): the **Chief of Staff** takes your request and hands it to the right specialist in the same thread. Repeatable orchestration belongs in a routine. Coordinating agents like Content Director still exist for jobs that need one tightly-orchestrated answer from several specialists at once — see [Best practices → Delegation & teams](/agents/best-practices#delegation-teams).
:::

::: warning Custom agents can't be team members
Custom agents are standalone by design; a coordinating agent's team is made of built-in specialists only.
:::

## Best practices

**Do**

* Start from the built-in agent whose model and defaults already fit the job — a specialist (Content Factory, Site Builder, Bulk Text Worker…) before a general-purpose one — and customize only what you must.
* Keep customizations light — a small set of changes is easier to reason about and survives future updates to the built-in defaults.
* To move an agent onto a specific or cheaper model, just change it in **Settings → Identity → Model** — no need to create a custom agent for that.
* When editing the **Skills** or **Labels** list, re-add everything you want to keep.

**Don't**

* Don't clear the Skills list down to one skill and assume the built-in ones are still attached — the list is what's attached.
* Don't build a custom agent just to switch models — the model picker is on every agent's Settings tab.
* Don't try to delete a built-in agent. To stop using one, **pause it** (Settings → Pause this agent) — that hides it and blocks new conversations while keeping existing ones readable. Only custom agents can be deleted.

## Test it

1. Open an agent and make a change on the **Settings** or **Instructions** tab (change the instructions, or toggle a tool on). Save, and confirm the **Customized** badge appears.
2. Send a message and verify the new behavior — the extra tool is available, or the new prompt voice shows up.
3. Click **Reset to default** and confirm every field reverts to how the agent ships and the Customized badge clears.
4. For a custom agent: create one with a specific model, run it once, and confirm the reply comes from the model you picked.
5. Change **Settings → Identity → Model** on a built-in agent, run once, and confirm the run is served by the new model — the model is a default you can move, not a lock.

## See also

* [The lobby](/agents/lobby) — the Chief of Staff front door where work starts
* [Your team & the agent library](/agents/library) — the roster, and adding/removing library agents
* [Build an agent](/agents/build-an-agent) — the end-to-end setup flow
* [Custom agents](/agents/custom-agents) — build your own from scratch
* [Tools & toggles](/agents/tools-and-toggles) — what each tool in "What it can do" lets the agent do
* [Models & reasoning](/agents/models-and-reasoning) — picking a model and reasoning effort
* [Skills](/agents/skills) — built-in vs. your own skills and attaching them
* [Tasks, crons & schedules](/agents/tasks-and-schedules) — scheduled runs per conversation
* [Routines](/agents/routines) — built-in and custom multi-step workflows agents run for you
* [Build routines with Claude or Codex](/guides/build-routines-with-claude) — configure, validate, preview, and stage your own workflow through the Agents MCP
* [Memory](/agents/memory) — conversation and long-term memory
* [MCP & connectors](/agents/mcp-and-connectors) — configure the Agents Platform or connect Product, Sales-Management, and UGC tools
* [Usage & billing](/agents/usage-and-billing) — the two buckets your spend draws from, reading the Balance card, and why agents stop