If you'd prefer to read this guide on notion, you can do so here.

✅ Overview

This setup enables fast prototyping, easy debugging, and scalable AI feature delivery.

1. 🧱 Define Your AI Task in Rightbrain

Rightbrain tasks are the core logic layer of your AI app.

What to do:

1. Go to Rightbrain

2. Create a new task

3. Fill in:

   • User Prompt with a task

     E.g.: "You are a contract analyst. Given a PDF, extract key legal clauses in JSON format."

   • Input variables

     E.g.: PDF file, plain text, image, or URL

   • Output format

     E.g.:

     {
       "clauses": [...],
       "summary": "..."
     }

🧪 Test

• Run sample inputs via the Rightbrain console

• Refine prompts or formatting as needed

• Ensure outputs are consistent and structured

2. 📦 Create an Agent Pack

This is your integration bundle — everything a dev or agent needs to integrate your task.

Your Agent Pack should include:

• API endpoint (from Rightbrain) - Get this from the Integrate tab

• Task name + description

• Input variables and output structure

• API key (Rightbrain lets you generate this instantly)

• Optional: human-readable guide to task behaviour

Suggested formats:

• Markdown (.md) for inline documentation

• .json where relevant to showcase output formatting

3. 🛠 Integrate in Replit

Replit is ideal for prototyping and deploying UI that connects to your core Rightbrain AI module.

Steps:

1. Create a new Replit app

2. Add your Agent Pack file

3. Add instructions on the functionality you’d like in the app

4. 🔍 Monitor + Debug

Rightbrain offers full monitoring and observability into LLM behaviour.

Use the Task Run logs to:

• Track task runs in real time

• View raw input and output

• Spot formatting issues, prompt failures, or authentication errors

This is useful during development and in production.

5. 🚀 Optimise and Iterate Safely

AI products evolve. Rightbrain makes that safe and trackable.

Features:

• Built-in versioning: Every change to your prompt or schema is a new revision

• Non-breaking updates: Replit integrations can lock to a task version

• Model switching: Try different providers (e.g. GPT-4, Claude, Gemini Flash)

Example use cases:

• Swap models to reduce cost or latency

• Improve output structure for easier rendering

• Add more edge case handling via prompt refinement

✅ Final Checklist

Before going live:

• [ ] Confirm API key is securely stored in Replit

• [ ] Test with varied input data

• [ ] Check Rightbrain logs to confirm runs

• [ ] Ensure UI handles all fields returned in output

• [ ] Optimise through updates to your prompt and model

🧠 Summary

Rightbrain handles the AI logic. Replit handles the interface. Together, they let you go from prompt → prototype in minutes.

Modular. Observable. Version-controlled.