Prompt Map Builder
A complete guide to building production-ready AI agent prompts — visually, without writing a single line of raw prompt text.
What is the Prompt Map Builder?
The Prompt Map Builder is a visual tool that helps you create system prompts for AI agents. Instead of staring at a blank text editor trying to write a wall of instructions, you work with a canvas of connected "nodes" — each one representing a piece of your prompt (role, tone, rules, examples, etc.).
Think of it as a flowchart for prompts. You pick a template, customize each section through guided forms, and the tool assembles a production-ready prompt for you in real-time.
You don't need to know what "system prompts," "tokens," or "temperature" mean. The Prompt Map Builder handles the technical side — you just fill in the blanks about your business and your agent.
Getting Started
Step 1: Open the Builder
Navigate to your Dashboard and click Prompt Map in the sidebar. You'll land on the template picker — a grid of pre-built starting points.
Step 2: Pick a Template
Click any template card to load it onto the canvas. Every field is already pre-filled with working defaults and examples — you can hit Export immediately and get a usable prompt, or customize everything.
Step 3: Edit the Nodes
Click into any node on the canvas to see its fields. Each field has a placeholder showing you what to type, helper text explaining why it matters, and sensible defaults you can keep or replace.
Step 4: Watch the Preview
The right panel shows your assembled prompt updating in real-time as you type. Every change you make in a node instantly appears in the preview. This is the actual text your AI agent will receive.
Step 5: Export
When you're happy, click Export to copy, download, or share your prompt. That's it.
Choosing a Template
Five templates are available, each pre-filled with different content:
Customer Support Agent
Pre-filled for e-commerce support. Includes return policies, shipping FAQ, escalation rules, example conversations, and opening/closing messages. Tone is set to friendly and empathetic.
Includes: 7 nodes: Role, Tone, Knowledge, Rules, Examples, Opening, Closing
Best for: Support teams, e-commerce, help desks
Marketing Content Agent
Built for social media and blog content. Includes brand voice guidelines, target audience info, creative writing examples, and tools like image generation and social scheduling.
Includes: 6 nodes: Role, Tone, Knowledge, Rules, Examples, Tools
Best for: Content creators, social media managers, marketing teams
Data Analyst Agent
For querying data and dashboards. Includes metric definitions, data source references, context variables (user name, role, date range), and tools like database queries and chart generation.
Includes: 7 nodes: Role, Tone, Knowledge, Rules, Examples, Tools, Context
Best for: Analytics teams, data-driven businesses
General Purpose Agent
Minimal defaults with lots of placeholder guidance. Good for experimenting or building something that doesn't fit the other categories.
Includes: 4 nodes: Role, Tone, Rules, Examples
Best for: Exploration, custom agents, learning
Blank Canvas
Starts with just a Role node. Add everything else from the palette. For users who want full control from scratch.
Includes: 1 node: Role
Best for: Power users, advanced builders
The Canvas
The canvas is your workspace — a zoomable, pannable surface where your prompt nodes live. Here's everything you can do:
Navigation
- Scroll to zoom in and out (0.2x to 2x)
- Click and drag the background to pan around
- Use the minimap in the bottom-left corner for orientation
- Click "Fit View" in the toolbar to see all nodes at once
Moving Nodes
- Click and drag any node to move it
- Nodes snap to a 20px grid for clean alignment
- Connections between nodes update automatically as you move them
Connecting Nodes
- Each node has a dot at the top (input) and bottom (output)
- Drag from one node's output dot to another node's input dot to create a connection
- Connections determine the order sections appear in your final prompt
- Delete connections by selecting the line and pressing Delete/Backspace
Toolbar (bottom center)
Keyboard Shortcuts
Adding Nodes from the Palette
The left sidebar shows all available node types. Click the + button next to any type to add it to your canvas. Most node types can only be added once (except Knowledge, which can be duplicated for multiple sources). Nodes already on the canvas show "Added" and are grayed out.
Node Types Explained
Every node has a colored header, a collapse/expand toggle, a status indicator (green checkmark = complete, yellow warning = using defaults, red X = empty), and a three-dot menu with Duplicate, Reset to Default, and Delete options. Each also has a lightbulb button that explains why that section matters.
Role
RequiredDefines who your agent is. This is the foundation of your prompt — it shapes everything else.
Tone & Style
Controls how the agent communicates — from casual to formal, short to detailed.
Knowledge
What the agent knows. This is where you paste your FAQ, product info, or company policies. Can be duplicated for multiple sources.
Rules
Guardrails that prevent the agent from going off-script. Pre-built rules organized by category, plus custom rules.
Examples
Most ImpactfulShow the agent what a good conversation looks like. This is the single most impactful section — it's worth more than 100 rules. The Quality Score weights it at 25 points (the highest).
Tools
External capabilities the agent has access to — like searching a database or creating tickets.
Memory
How the agent handles conversation context and what it should remember.
Opening Message
How the agent introduces itself and starts conversations.
Closing Behavior
How the agent wraps up conversations.
Context Variables
Dynamic variables injected at runtime — like the user's name or plan tier. Advanced feature.
Live Preview & Quality Score
The Preview Panel
The right sidebar shows your fully assembled prompt, updating in real-time as you edit any node. The prompt is assembled by walking the node connections (or using default section order if there are no connections): Role, Tone, Knowledge, Rules, Examples, Tools, Memory, Opening, Closing, Context.
- Copy All button copies the entire prompt to your clipboard
- Word count and estimated token count shown at the bottom
- Edit as Text mode (pencil icon) lets you manually edit the output — though changes won't sync back to nodes
Quality Score
The Prompt Strength score (0-100) grades your prompt based on completeness:
The bar is colored red (<40), yellow (40-70), or green (>70). Below the score, you'll see the most impactful next step: "Add 2 example conversations to jump from 75 to 100."
Exporting Your Prompt
Click the Export button in the preview panel (requires Role + at least one other node). A modal opens with three tabs:
Copy
Toggle between Markdown (with headers) and Plain Text (headers stripped). Click Copy to Clipboard — paste directly into ChatGPT, Claude, your API call, or wherever you use system prompts.
Download
- .md — Markdown file with section headers and formatting (recommended)
- .json — Structured JSON with metadata (name, type, version, timestamp, strength score) plus the full prompt. Great for programmatic use.
- .txt — Plain text, no formatting
Share
Generate a unique shareable link (requires sign-in). Anyone with the link can view your prompt map in read-only mode, see all the nodes on the canvas, copy the assembled prompt, and build their own version. Social sharing buttons for X and LinkedIn are included.
Mobile Usage
On screens smaller than 768px, the canvas switches to an accordion layout — same nodes, same fields, same editing, just stacked vertically instead of on a canvas. Each section is expandable/collapsible. The Quality Score and prompt preview are available at the bottom.
Tips & Best Practices
Start with a template, always
Even if you plan to customize everything, start from a template. The pre-filled defaults give you a working prompt in seconds, and you can see what "good" looks like before making changes.
Examples are your secret weapon
The Examples node is weighted at 25 points in the Quality Score — more than any other section. Two or three good example conversations will improve your agent's responses more than pages of rules. Show, don't tell.
Add a bad example too
A single "don't respond like this" example with a clear reason teaches the AI what to avoid. It's especially powerful for preventing dismissive or vague responses.
Use the Knowledge node for specifics
Don't put company info in the Role node. The Knowledge node is designed for detailed content — paste your entire FAQ, return policy, or product catalog there. You can duplicate the Knowledge node for multiple sources.
Define escalation triggers early
Decide upfront when the agent should stop trying and hand off to a human. "Billing disputes, legal questions, angry customers" is a good starting set. This prevents your agent from stumbling through conversations it can't handle.
Use Auto-Layout after adding nodes
After adding or rearranging nodes, click the Auto-Layout button (circular arrows icon) to automatically arrange everything in a clean top-to-bottom layout. Then hit Fit View to center it all.
Export as JSON for production use
If you're building an app, download as .json. It includes structured metadata, the full prompt, and all section data — ready to parse programmatically.
Aim for 80+ on the Quality Score
A score above 80 means you have a well-rounded prompt. The suggestions below the score bar tell you exactly what to add next for the biggest impact.
Ready to build your first agent prompt?
Pick a template, customize the nodes, and export a production-ready system prompt in under 5 minutes.
Open Prompt Map Builder