Skip to content
AI Agents

Agents that use tools, not just talk

An agent is a reusable assistant: a bound model, a defining persona, and a curated set of tools — all working together. Give it a goal and watch it reason, call the right tools, read the results, and deliver a complete answer.

The concept

What is an agent?

An agent is a three-part bundle: a provider and model, a system prompt that defines its persona and instructions, and a curated set of tools it is allowed to call. Once configured, it works autonomously — deciding when to reach for a tool and when to respond.

Bound model & persona

Each agent is locked to a specific provider and model — OpenAI, Anthropic, Gemini, Ollama, and more. A system prompt defines its role, tone, and constraints so it behaves consistently every time.

  • Choose any connected provider and model
  • Version-controlled system prompts
  • Persona stays consistent across all conversations

Enabled tools

You decide which tools the agent may call — built-in utilities like web search, or your own HTTP endpoints. The agent only has access to the tools you explicitly enable, nothing else.

  • Granular per-agent tool permissions
  • Mix built-in and dynamic HTTP tools
  • Every tool call is visible in the thread

Autonomous tool-calling

The model decides when a tool is needed, constructs the call, reads the result, and continues reasoning — all without human intervention. Every step streams live in the chat thread.

  • Zero-shot tool selection by the model
  • Multi-step tool chains in a single turn
  • Full call-and-result trail visible inline

Step by step

How an agent works

From the moment you send a message to the final streamed response, here is what happens inside every agent turn.

  1. User sends a prompt

    The conversation turn is forwarded to the bound model with its full history and system prompt.

  2. Model reasons

    The model reads the system prompt and conversation history, then decides what to do next — answer directly or call a tool.

  3. Tool call issued

    When more information is needed, the model emits a structured tool-call (e.g. web_search or http_request).

  4. Result returned

    The platform executes the tool, captures the result, and streams it back into the model's context — all visible in the thread.

  5. Model responds

    With full context including tool results, the model produces its final answer — streamed live, token by token.

Every tool call and its result appear inline in the chat thread, streamed in real time.

Use cases

What can you build?

Agents are general-purpose — here are a few of the most common starting points, each deployable in minutes.

Research assistant

Give the agent web search and URL-reading tools. Ask a question and it scouts multiple sources, cross-references them, and returns a cited markdown report — in a single turn.

  • Automatic multi-source cross-referencing
  • Inline citations in the response
  • Configurable search depth and source limit

Support triage

Bind the agent to your knowledge-base API and ticket system. It classifies incoming requests, fetches the relevant docs, drafts a reply, and escalates when confidence is low.

  • Classifies and routes issues automatically
  • Pulls from live knowledge-base via HTTP tool
  • Escalation trigger on low-confidence answers

Data lookup & internal API

Expose your internal REST API as an HTTP tool. The agent queries it on demand — fetching records, aggregating results, and formatting them for the user — all inside the chat.

  • No custom integration code required
  • Auth headers stored securely per tool
  • Structured results rendered as tables

Content drafting

Pair a writing-focused system prompt with a web search tool. The agent researches the topic, outlines the piece, and drafts each section — ready for your review and iteration.

  • Research-then-write loop in one agent
  • Consistent tone enforced via system prompt
  • Iterative revision supported in the same thread

Agent builder

Build an agent in minutes

Open the agent builder inside the app, give it a name, pick your provider and model, write a system prompt, and select the tools it may use. Hit save — your agent is live instantly, no deployment required.

The agent builder: name, model selector, system-prompt editor, and tool-permission toggles in one panel.

Share agents with your team

Once an agent is saved you can publish it to your team workspace. Every member can use it in their own chats and workflows — and you control who can edit the definition. Learn about team collaboration →

Start building with The Chat Agent

Connect your favourite AI providers, add tools and agents, and ship in minutes. Free to start — no credit card required.