The product

The Builder

The Builder is Forgent's main screen: a two-column chat — the conversation on the left, the blueprint being built live on the right. Five questions, an agent ready to deploy.

Free to build and deployThe Builder assistant is funded by the platform, so designing an agent costs you nothing and needs no API key. Deploying is free too — Forgent runs your agents for you, with nothing to plug in (see Deploy an agent).

The two-column interface

On the left, a chat. On the right, your agent's blueprint filling in field by field as you chat. You see in real time the impact of each answer on the final configuration.

On mobile, the same thing but in tabbed mode: a Conversation / Preview toggle at the top to switch between the two panels.

The 5 steps

The Builder always follows the same structure to stay predictable. The progress bar at the top shows where you are.

1. Description

You explain in plain English what your agent should do. The more precise you are about the trigger (when the agent should run) and the deliverable (what it should produce), the better. Avoid technical jargon — talk to it like an intern.

2. Model

Your agent runs on Claude Fable 5, Anthropic's most capable model. See Claude Fable 5 for the details.

3. Tools

The agent is given access to a set of tools: reading / writing files, running bash, web search, MCP servers (Notion, Gmail, GitHub...). The Builder suggests the relevant tools based on your description, and you approve or modify them.

4. Tone

You set your agent's tone — clear and factual, warm, formal, technical. This choice shapes the system prompt but also the format of the outputs.

5. Summary

The Builder shows the complete blueprint, gives you a summary in plain English of what the agent will do, and offers to deploy.

The system prompt is invisibleYou won't see it anywhere in the Builder, and that's intentional. It's generated and optimized in the background for Claude by Claude. If you really want to read it, it appears on the Deploy page before going to production.

Best practices for describing an agent

  • State the trigger: "every Monday at 9am", "when an email arrives", "on demand".
  • State the deliverable: "an email with a 5-bullet-point summary", "a CSV file added to my Drive".
  • Limit the scope: an agent that does ONE thing well beats an agent that tries to do ten things mediocrely.
  • Give an example if the task is subtle: an example of expected input and output is worth 1000 words of explanation.

Resuming a conversation

Your in-progress blueprint is auto-saved to your account as you work (debounced, every second or two). Close the tab and come back later — even on another device — and your draft is right where you left it. A local copy is also kept in the browser for zero-latency editing.