> ## Documentation Index
> Fetch the complete documentation index at: https://docs.collivo.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Agent Modes & Controls

> Choose a mode, and control what the agent sees and how it behaves.

export const SharedFolder = () => <span style={{
  whiteSpace: "nowrap"
}}>
    <svg width="17" height="17" viewBox="0 0 24 24" fill="currentColor" fillOpacity="0.35" stroke="currentColor" strokeOpacity="0.7" strokeWidth="1.25" strokeLinecap="round" strokeLinejoin="round" style={{
  color: "light-dark(#2E7D32, #4CAF50)",
  display: "inline-block",
  verticalAlign: "-0.2em",
  marginRight: "0.3em"
}} aria-hidden="true">
      <path d="M20 20a2 2 0 0 0 2-2V8a2 2 0 0 0-2-2h-7.9a2 2 0 0 1-1.69-.9L9.6 3.9A2 2 0 0 0 7.93 3H4a2 2 0 0 0-2 2v13a2 2 0 0 0 2 2Z" />
      <path d="M2 10h20" />
    </svg>
    <strong>Shared</strong>
  </span>;

export const PrivateFolder = () => <span style={{
  whiteSpace: "nowrap"
}}>
    <svg width="17" height="17" viewBox="0 0 24 24" fill="currentColor" fillOpacity="0.35" stroke="currentColor" strokeOpacity="0.7" strokeWidth="1.25" strokeLinecap="round" strokeLinejoin="round" style={{
  color: "inherit",
  display: "inline-block",
  verticalAlign: "-0.2em",
  marginRight: "0.3em"
}} aria-hidden="true">
      <path d="M20 20a2 2 0 0 0 2-2V8a2 2 0 0 0-2-2h-7.9a2 2 0 0 1-1.69-.9L9.6 3.9A2 2 0 0 0 7.93 3H4a2 2 0 0 0-2 2v13a2 2 0 0 0 2 2Z" />
      <path d="M2 10h20" />
    </svg>
    <strong>Private</strong>
  </span>;

The AI you chat with is an agent. You shape how it works from the **chat toolbar** below the message box — which controls both what the agent can see and how it behaves — starting with the **thinking mode**.

## Thinking modes

The thinking mode changes how the agent reasons internally before it answers. Collivo offers three, chosen from the thinking-mode selector in the toolbar.

| Mode | How it works | Best for |
| - | - | - |
| **Quick** | Fast, direct responses. Usually the best place to start. | Simple questions and answers, small edits, summaries, quick drafts, finding files |
| **Expert** | Works through a problem step by step before answering. It takes more time to check and refine its answer, so it can go deeper on complex work. | Research, analysis, comparing options, longer multi-step tasks |
| **Expert Panel** | Multiple specialized agents collaborate, debate, and synthesize a single answer. | Strategy, high-stakes decisions, evaluating tradeoffs, structured critique |

Start with **Quick**, and move to **Expert** or **Expert Panel** as the task gets more ambiguous or higher-stakes. Higher modes take longer, so match the mode to the task.

## Control what the agent can see

The agent works within your current working folder (see [Chat basics](/ai-and-agents/chat-basics)). The <Icon icon="paperclip" color="#1F7A5A" /> **Context** control shows the files currently attached to the conversation and lets you add specific ones, so the agent focuses on exactly what the task needs. The number on the button is how many files are attached.

You can also paste a screenshot straight into the message box — Collivo uploads it to the project and tags it in your message, so the agent can see it right away.

To bring new files into a project, use the **Files** pane — see [Upload and organize files](/project-basics/upload-and-organize-files). (In a standalone quick chat, an **Add files** button in the toolbar uploads directly.)

## Control how the agent behaves

The rest of the toolbar shapes how the agent works on your task:

* <Icon icon="drama" color="#1F7A5A" /> **Persona** — defines the agent's identity: who it is. Apply one so the agent takes on a particular role, expertise, or perspective.
* <Icon icon="biceps-flexed" color="#1F7A5A" /> **Skills** — impart specific capabilities to the agent. The agent can draw on a skill on its own whenever it is relevant; use the skill menu only when you want to require it to follow a particular skill's guidance for your next message.
* **Model** — chooses which AI model powers the agent's reasoning — the brain behind it. In Expert Panel this is set automatically (Auto / Smart Selection).
* <Icon icon="list-checks" color="#1F7A5A" /> **Plan mode** — controls whether the agent plans before it acts. It appears only in **Quick** mode, where you can turn planning off for a faster, more direct response. **Expert** and **Expert Panel** always plan, and it cannot be turned off.
* **Tools** — turn individual tools on or off, such as **Write** (create or change content) and **Web Search**. Pin the tools you use often to the toolbar; find the rest under **More tools**. Some tools are always on.

## Approvals

By default, the approval guardrail is on: the agent asks before actions that change content. For supported personal work, you can use the approval shield <Icon icon="shield-check" color="#1F7A5A" /> to relax that guardrail and speed up turns. The control is available when you are working in your <PrivateFolder /> area, or in a folder that contains it. When the guardrail is relaxed, the agent can act without asking on files that are both inside your <PrivateFolder /> area and inside the current working folder. Anything else still needs your approval, including content in <SharedFolder />, external tools and connected accounts, and actions that do not target a specific file, such as running code. See [Tool approvals](/ai-and-agents/tool-approvals).

## Review the output

Whatever mode and settings you use, the output is assistance, not final authority. Review important results before relying on them. See [AI permissions, context, and privacy](/sharing-and-security/ai-privacy-limitations).

## Related articles

* [Chat basics](/ai-and-agents/chat-basics)
* [Tool approvals](/ai-and-agents/tool-approvals)
* [AI permissions, context, and privacy](/sharing-and-security/ai-privacy-limitations)


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.