Agents take The Chat Agent beyond question-and-answer chat. An agent has a persona, a goal, and a set of tools it can call — so instead of answering from training data alone, it can search the web, run code, generate images, or reach out to any HTTP endpoint you connect. This guide walks through creating your first agent, choosing the right tools, and refining the experience as you go.

Creating an agent

Open the Agents section from the sidebar and click New agent. The agent builder gives you three things to fill in:

Model. Choose which AI model powers the agent. You can pick any model you have connected — OpenAI, Anthropic, Google, or a local endpoint. Different models suit different tasks; a reasoning-focused model works well for research, while a faster model is better for quick lookups.
Persona (system prompt). Write a short paragraph describing what the agent is for, how it should behave, and what it should focus on. The more specific you are here, the more consistent and useful the agent becomes. For example: "You are a concise research assistant. Search the web before answering any factual question. Always include the source URL."
Tools. Check the boxes next to the tools this agent is allowed to use. Enabling only the tools the agent actually needs keeps its behaviour predictable and focused.

Hit Save, and the agent appears in your agent list. You can start chatting with it immediately.

Built-in tools

The Chat Agent ships four built-in tools that any agent can use with a single checkbox:

Web search — performs a live web search and returns titles, URLs, and snippets. Useful for any agent that needs current information.
Fetch page — reads the plain-text content of a public web page. Pair it with web search so the agent can follow up a result by reading the full article or document.
Code execution — runs Python in an isolated sandbox. Ideal for calculations, data manipulation, and generating simple charts.
Image generation — creates an image from a text description using your connected image provider (DALL·E, Stable Diffusion, etc.).

Enable only the tools relevant to what the agent is supposed to do. A research assistant probably needs web search and fetch page; it probably does not need image generation.

Dynamic tools

Beyond the built-ins, you can connect any HTTP endpoint as a custom tool. Head to Settings → Tools and click Add tool. Give the tool a name, write a short description of what it does (the model reads this to decide when to call it), point it at an endpoint, and define the parameters it accepts. Once saved, the tool appears in the tool picker inside every agent builder — check the box to enable it for a specific agent.

Dynamic tools are useful for connecting to your own services: a knowledge base, a calendar API, a project management tool — anything with an HTTP interface. The platform handles sending requests, injecting parameters, and passing the response back to the model.

Watching tool calls in the conversation

When an agent uses a tool, the conversation shows the work, not just the final answer. Each tool call appears inline in the thread: you can see which tool was called, what arguments were passed, and what came back. The model's reasoning step connects the tool result to the final response.

This transparency makes it easy to spot when an agent is doing the right thing and easy to diagnose when it is not. If the search results look off, or the agent is calling a tool it should not be, you can go back into the agent builder and adjust the persona or the tool selection.

Iterating on the persona

The system prompt is the highest-leverage thing you can adjust. A vague prompt produces a generic agent; a specific, well-scoped prompt produces one that is genuinely useful every day. A few things that help:

Tell the agent what format you want answers in — bullet points, short paragraphs, a table.
Tell it what to do when it is uncertain — say so, or search before guessing.
Give it a clear focus. An agent scoped to one domain stays useful longer than one asked to do everything.

Save a new version, chat with it, and compare. If an edit makes things worse, you can open the agent builder and revert your changes before saving.

Keep reading

The following resources go deeper into agents and tools inside The Chat Agent:

AI Agents overview — the full feature reference, including multi-agent workflows.
Dynamic tools — a step-by-step walkthrough for connecting your first custom HTTP tool.

Getting the most from agents & tools