---
title: "Respan Agent"
subtitle: "Ask questions about your LLM traffic, and let the agent set up prompts, monitors, and evaluations for you."
---

The Respan agent is a chat assistant built into the platform. It reads your logs, traces, and metrics to answer questions, and it can create or change prompts, datasets, evaluators, experiments, and monitors for you.

| Goal | Try asking |
| --- | --- |
| Investigate a spike or failure | "Why did errors spike yesterday?" |
| Debug a trace | "Explain why trace [trace_id] is slow" |
| Find requests by what they say | "Find logs that mention 'rate limit exceeded' in the last 24 hours" |
| Write a prompt | "Create a customer support prompt" |
| Get alerted | "Create a monitor when cost exceeds $100/hour" |
| Evaluate outputs | "Build a hallucination evaluator" |

## Open the agent

- **Chat** in the sidebar opens the full-page chat. **New chat** and your past chats are listed under it.
- **Ask Respan agent**, at the bottom right of any page, opens a chat panel over that page. On pages such as Logs and Dashboard, a **Current page** chip shares what you're looking at with the agent; click its **×** to leave it out. Use the panel's header to switch chats, start a new one, manage skills, dock the panel beside the page, or close it.
- **Search** (⌘K): type your question and pick **Ask Respan agent** at the bottom of the results. Chat opens with your question filled in.

<img src={"/docs/images/use-ai/agent-panel-light.png"} width="400" className="block dark:hidden" alt="The Respan AI chat panel opened from Ask Respan agent, with saved skills, starter prompts, and a message box with the Current page chip" />
<img src={"/docs/images/use-ai/agent-panel-dark.png"} width="400" className="hidden dark:block" alt="The Respan AI chat panel opened from Ask Respan agent, with saved skills, starter prompts, and a message box with the Current page chip" />

## Ask a question

<Steps>
<Step title="Describe what you want">

Type a question, or pick a starter prompt or skill. Include anything the agent can't guess, such as a trace ID, a time range, or a threshold. Press Enter to send.

</Step>
<Step title="Read the answer">

As it works, the agent lists each step it runs, like "Error count by model". Click a step to see what it queried and what came back. Charts in the answer are drawn from your data.

<img src={"/docs/images/use-ai/agent-answer-light.png"} className="block dark:hidden" alt="The agent answering 'Why did errors spike yesterday?' with an errors-per-hour chart by model, the cause, a fix, and suggested follow-up questions" />
<img src={"/docs/images/use-ai/agent-answer-dark.png"} className="hidden dark:block" alt="The agent answering 'Why did errors spike yesterday?' with an errors-per-hour chart by model, the cause, a fix, and suggested follow-up questions" />

Click a suggestion under the answer to ask it next.

</Step>
<Step title="Check what it changed">

When the agent creates or edits something, the answer ends with an **Agent updated** button, such as **Agent updated prompt**, that opens the result. The changes are real, so review them before you rely on them.

</Step>
</Steps>

## Add context

Click **+** in the message box, or type `@`, to give the agent more to work with.

<img src={"/docs/images/use-ai/agent-context-menu-light.png"} width="240" className="block dark:hidden" alt="The + menu in the message box: Add images or documents, Saved skills, and Reference context" />
<img src={"/docs/images/use-ai/agent-context-menu-dark.png"} width="240" className="hidden dark:block" alt="The + menu in the message box: Add images or documents, Saved skills, and Reference context" />

| Option | Adds |
| --- | --- |
| **Add images or documents** | Up to 5 files per message, 20 MB each: PNG, JPEG, GIF, WebP, PDF, TXT, Markdown, CSV, or JSON |
| **Saved skills** | Up to 5 saved skills. With text in the box, **Save current prompt as skill** saves it as a new skill. |
| **Reference context** | Up to 5 prompts, graders, or monitors from your project, or the **Current page** if you removed it |

## Save a skill

A skill is a set of instructions that anyone in your organization can apply to a chat. **Root cause analysis** is built in.

<Steps>
<Step title="Create the skill">

On the **Chat** page, click **Skills** at the top right, or click the skills icon in the header of the **Ask Respan agent** panel. In **Agent skills**, click **New skill**.

</Step>
<Step title="Write the instructions">

Enter a **Name** and a **Description**, then write the instructions the agent should follow. Use **+** or `@` to attach a prompt, grader, or monitor the skill depends on.

</Step>
<Step title="Save and use it">

Click **Create skill**. It shows up under **Skills** on the Chat page and in **+** › **Saved skills**.

</Step>
</Steps>

## Credits

The agent runs on your Respan credits. If your balance is empty, the chat shows an error with a **Buy credits** button instead of an answer. See [Billing and credits](/docs/documentation/admin/credits-billing).