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

# Configure an Agent

> Edit your custom agent's model, system prompt, tools, connectors, and other settings.

You can edit any aspect of a custom agent after creating it — change its model, refine the system prompt, adjust which tools it has access to, or add new connectors.

## Accessing Agent Settings

1. Go to `/agents` from the topbar
2. Find the agent you want to configure
3. Click the **Settings** or **Edit** button on the agent's card

The agent editor opens with all configurable fields.

## What You Can Configure

### Name and Description

Update the agent's display name or description at any time. The name appears in the agent list and at the top of every chat with this agent.

### Avatar / Icon

Change the agent's visual identity — useful when you have many agents and want to distinguish them at a glance.

### Base Model

Change which AI model powers the agent. You can switch models without changing anything else about the agent.

**Choosing the right model:**

| Model type             | Best for                                                   |
| ---------------------- | ---------------------------------------------------------- |
| **Fast/lightweight**   | Quick responses, simple tasks, high-volume use             |
| **Reasoning/powerful** | Complex analysis, debugging, multi-step logic, code review |
| **Balanced**           | General-purpose agents that need both speed and capability |
| **Multimodal**         | Agents that work with images or visual content             |

Changing the model affects all future chats — previous chats retain the model that was used at the time.

### System Prompt

The system prompt is the most impactful thing you can change. Small wording improvements often produce significant behavioral changes.

**Common refinements:**

* **Too verbose?** → Ask the agent to be more concise; specify word limits
* **Wrong tone?** → Add explicit tone guidance ("Be direct and technical, not conversational")
* **Inconsistent format?** → Specify the exact format: "Always structure responses as: Summary → Details → Next steps"
* **Scope creep?** → Add constraints: "Only help with TypeScript questions; redirect anything off-topic"
* **Missing context?** → Add background your agent needs to give relevant answers

### Enabled Tools

Toggle which tools this agent has access to:

| Tool              | What it enables                                                           |
| ----------------- | ------------------------------------------------------------------------- |
| **Web search**    | Agent can search the web for current information                          |
| **Deep research** | Agent can run extended, multi-source research sessions                    |
| **Canvas**        | Agent can create and edit documents and structured outputs in canvas mode |
| **File uploads**  | User can attach files to chats with this agent                            |
| **Voice**         | Agent can respond with voice output                                       |

### Connected Apps (Connectors)

Select which of your connected OAuth apps this agent can access. Only apps already connected in **Settings → Connectors** appear here.

Changes take effect immediately for new chats. The agent can only use apps you explicitly select for it.

## Testing After Changes

After editing an agent, always test it:

1. Save your changes
2. Start a new chat with the agent
3. Ask it a task representative of its purpose
4. Evaluate: Is the tone right? Is the format right? Does it follow your instructions?

**If the system prompt isn't working:**

* Be more explicit — vague instructions produce inconsistent behavior
* Use numbered rules instead of prose for critical constraints
* Add a short example of ideal output at the end of your prompt
* Test with edge cases (unusual questions, off-topic requests) to verify constraints hold

**Iteration is normal.** Most agents need 3–5 rounds of prompt refinement before they consistently behave exactly as intended.

## Best Practices

* **Make incremental changes.** Change one thing at a time (model, then prompt, then tools) so you can isolate what's affecting behavior.
* **Keep a backup of your system prompt.** Before making major changes, copy your current prompt somewhere so you can revert if the new version is worse.
* **Use specific, testable success criteria.** Before editing, define what "better" looks like — then test against that definition after each change.