The product

The Blueprint

A blueprintis the complete spec sheet of your agent: its name, the Claude model it uses, the tools it has access to, the instructions it must follow. It's what the Builderconstructs for you throughout the conversation, and it's what Forgent sends to Anthropic Managed Agents at deployment time.

Why this word?It's the official term used by Anthropic in their API. We use it to stay consistent with their documentation, but you can simply think "my agent's spec sheet" or "my agent's plan" — it's exactly the same thing.

The essential fields

Name

Short (max 60 characters), descriptive, in any language. Used to find the agent in My Agents and shown in the Anthropic dashboard.

Description

One or two sentences summarizing what the agent does. It's the summary shown in the list — not the system prompt.

Model

The Claude model that runs the agent: claude-fable-5, Anthropic's most capable model. See Claude Fable 5 for the details.

Tools

The list of capabilities the agent has access to. A few standard tools:

  • read, write — reading / writing files
  • bash — running sandboxed shell commands
  • web_search, web_fetch — searching and fetching web pages
  • MCP: notion, MCP: gmail, etc. — connectors to your external tools (see MCP & tools)

Tone

Influences both the voice of the agent (formal, warm, technical) and the format of its outputs (bullet points, tables, running prose). A subtle setting but one that changes the experience of a deliverable a lot.

System prompt

The heart of the blueprint: the complete instructions given to Claude. It's generated automatically by the Builder from your answers. You can view it before deployment and edit it manually if you know what you're doing — but in general, the generated prompt is better written than a human prompt because it's been optimized for Claude by Claude.

JSON format

Here's an example of a serialized blueprint. It's what you find in My Agents under View JSON, and it's portable:

json
{
  "name": "Weekly competitive monitoring",
  "description": "Watches 5 competitor sites and sends a summary every Monday.",
  "model": "claude-fable-5",
  "tools": ["web_fetch", "web_search", "write"],
  "tone": "Clear and factual",
  "system": "You are a competitive monitoring agent. Every Monday at 9am, you visit the 5 sites listed below, you identify what's new (new pages, blog posts, pricing changes) and you produce a structured markdown report...",
  "schedule": "weekly",
  "outputs": ["markdown"]
}
Why it's portableThis JSON is exactly what Anthropic Managed Agents accepts as input. If tomorrow you want to leave Forgent and hit the Anthropic API directly, copy-paste is all it takes. We don't lock you in.

Versions and history

Forgent keeps a history of the latest snapshots of your blueprint (persisted server-side, so it follows you across devices) and lets you restore an old version in the Builder. See Versions & detailed logs for the details.