Customisable AI assistants, configured not coded

An agent is an AI assistant you shape: its own model, its own role, its own knowledge, and only the tools you choose to give it. Build one in a dialog, not a sprint.

All key features

Running

Correct it once, in plain English

This is the loop the whole product is built around. The agent acts, you see every step it took, and you say what should have happened instead — that sentence becomes a line in the agent's system prompt, so the fix holds from the next run onward.

Tool calls shown as they happen
Corrections written as system-prompt edits
The updated prompt applies to every future run

Refunds agent

Stripe · Postgres · Knowledge base

Live

Observability

TreeRaw JSON1f86ff405-1ffe6fb34432

Waiting for the next run…

Every agent starts with a role

Give the agent a name, the role it plays, and a description of what it does, then pick the model that powers it. That is the whole setup: no code, no deployment step.

Name, role, and description
Choose the model per agent
Active the moment it is created

The Create New Agent dialog, with fields for name, role, description, and model

Every agent, in one place

The agents page lists everything your organisation has built: the role each one plays, what it does, whether it is active, and when it was last used.

Role and description at a glance
Active status per agent
Sorted by last used

The agent management page listing agents with their role, description, status, and last-used time

Built-in tools, switched on per agent

Four pre-built tools ship with the platform. Enable only the ones a given agent should have. Capability is a decision you make per agent, never a default.

Chart Renderer for visualisations
Code Interpreter, sandboxed
Browser for web automation
Document Generator: PDF, DOCX, Excel

The agent's Tools tab showing Chart Renderer, Code Interpreter, Browser, and Document Generator, each with an enable toggle

Ground it in your knowledge

An agent inherits the knowledge base of the project it is assigned to, and you can attach more. Each source is toggled on or off independently.

Project knowledge base, auto-assigned
Attach additional knowledge bases
Toggle each source on or off

The agent's Knowledge Base tab, showing an auto-assigned project knowledge base and additional attached sources

Connect a server, then pick the tools

Point an agent at a Model Context Protocol server and it discovers the tools that server exposes. You decide which of them the agent may actually call.

Server health at a glance
Enable discovered tools one by one
Custom tools and integrations

An agent connected to an MCP server, listing its seven available tools with individual enable toggles

Chat with your agents

Talk to any agent from the app. Tool calls appear inline as the agent makes them, so you can see the work rather than just the answer.

Multiple conversations per agent
Switch agent mid-session
Tool calls shown inline
Attach files to a message

A chat with the Content Writer agent, showing an inline completed tool call followed by the drafted article

NextToolsTools are how an agent goes beyond answering questions. Send a prompt, and the agent decides whether a tool is needed, runs it, and uses the result in its response.

See Agents in action

Spin up your first agent, or walk through Chocolate Factory with our team on a live demo.