Salesforce AI Agent Framework

Enterprise-Grade AI Platform for Salesforce

Build intelligent AI agents powered by Large Language Models that seamlessly integrate with your Salesforce environment. Designed for security, scalability, and ease of use.

---

� Documentation

View Full Documentation →

• Getting Started Guide
• Configuration Reference
• Standard Actions
• Developer Guide - Custom actions & context providers
• Security Guide
• Troubleshooting

---

💖 Support This Project

AI Agent Studio is free and open-source. If you find it useful, consider supporting ongoing development.

☕ GitHub Sponsors | Buy Me a Coffee

---

⚠️ Repository Notice

This repository contains the core AI Agent Framework only. The aiAgentStudioAddons folder contains proprietary extensions not included in the open-source release.

The core framework in force-app provides all fundamental capabilities for building conversational AI agents with OpenAI.

---

🎯 What is This?

Create AI-powered assistants that can:

• 💬 Chat naturally with users and remember conversation context
• 🔍 Search and retrieve Salesforce data intelligently
• ✏️ Create and update records based on user requests
• 🔄 Execute workflows with multiple steps automatically
• 🔒 Respect permissions - agents only access what users can access
• 🎯 Work with multiple AI providers - OpenAI included, extensible for others

---

💼 Use Cases

Customer Support

Deploy AI copilots that help support agents resolve cases faster by automatically searching knowledge bases, pulling customer history, and suggesting solutions - all while respecting your existing security model.

Sales Enablement

Give sales reps an intelligent assistant that can find accounts, surface open opportunities, create follow-up tasks, and provide real-time insights during customer conversations.

Operations Automation

Build workflow agents that handle multi-step processes like lead qualification, case routing, or approval workflows - combining multiple AI-powered decisions into automated pipelines.

Self-Service Portals

Embed conversational agents in Experience Cloud to let customers check order status, create support cases, or find answers from your knowledge base without waiting for human agents.

---

🔄 How It Works

flowchart LR
    subgraph Input
        A[👤 User Message]
    end

    subgraph Framework
        B[🎭 Orchestrator]
        C[🧠 LLM]
        D[🔧 Tools]
    end

    subgraph Salesforce
        E[📊 Data]
        F[📝 Context]
        G[💾 Memory]
    end

    A --> B
    B --> C
    C --> D
    D --> E
    F --> B
    G --> C
    D --> H[✅ Response]

Loading

1. User sends a message through the chat component or API
2. Context is gathered from the current record, user profile, and related data
3. LLM analyzes the request with full conversation history
4. Tools execute Salesforce operations (query, create, update, post)
5. Response is delivered back to the user with full audit trail

All operations run asynchronously using Platform Events or Queueables, ensuring scalability for enterprise workloads.

---

✨ Key Features

Feature	Description
Three Agent Types	Conversational (chat), Function (single-task), Workflow (multi-step)
Smart Memory	Buffer window and summary-based conversation history
Built-in Security	Automatic CRUD, FLS, and sharing rule enforcement
Standard Actions	Create, update, query records, post to Chatter, execute Flows
Extensible	Custom actions, context providers, LLM adapters, memory managers
Observability	Full logging of LLM interactions, tool executions, and token usage
Async Processing	Platform Events (high concurrency) or Queueables (debugging)

---

🏆 Why This Framework?

Challenge	How We Solve It
Security concerns with AI	Runs in user context with automatic CRUD/FLS enforcement. No privilege escalation. Full audit trail.
Integration complexity	Native Salesforce - no external servers, middleware, or data sync. Works with your existing org.
Vendor lock-in	Bring your own LLM. OpenAI included, easily add Claude, Gemini, or others via adapter pattern.
Scalability	Async processing handles thousands of concurrent conversations. Choose Platform Events or Queueables.
Customization needs	Extensible architecture with interfaces for custom actions, context providers, and memory strategies.
Governance & compliance	Every interaction logged to AgentDecisionStep__c. See exactly what the AI decided and why.

---

🚀 Quick Start

Prerequisites

• Salesforce org (Sandbox recommended)
• System Administrator access
• OpenAI API key

Installation

Option 1: CumulusCI (Recommended for Scratch Orgs)

If you have CumulusCI set up:

cci flow run dev_org --org dev

This single command:

• Creates a scratch org with the framework deployed
• Deploys seed data and sample configurations
• Assigns required permission sets (AIAgentStudioConfigurator, AIAgentStudioEndUser)
• Enables Knowledge user and assigns KnowledgeDemo permission set
• Creates comprehensive sample data (agents, capabilities, test records)
• Sets up a Connected App for API access

Option 2: Salesforce CLI

sf project deploy start -d force-app/main/default -o your-org-alias

Setup

1. Create Named Credential for OpenAI API (https://api.openai.com)
2. Create LLM Configuration with your Named Credential and model settings
3. Create AI Agent Definition with identity/instruction prompts
4. Add Capabilities (tools) the agent can use
5. Add Chat Component to a Lightning page

👉 Full Getting Started Guide →

---

🏗️ Architecture

Core Components:

• AgentExecutionService - Main entry point for starting agent executions
• ConversationalOrchestrator - Manages multi-turn chat conversations
• LLMInteractionService - Handles communication with AI providers
• CapabilityExecutionService - Executes tools/actions securely
• OpenAIProviderAdapter - OpenAI/Azure OpenAI integration

Extension Points:

• IAgentAction - Build custom actions for any business logic
• IAgentContextProvider - Supply dynamic context to agents
• ILLMProviderAdapter - Add support for additional AI providers
• IMemoryManager - Implement custom conversation memory strategies

👉 Architecture Details → | Developer Guide →

---

⚠️ Important Notes

• Use at your own risk - Test thoroughly in sandbox before production
• AI content verification - LLMs can hallucinate; review automated actions
• Data privacy - User inputs are sent to external AI providers
• Cost awareness - Monitor token consumption; set appropriate history limits

---

📄 License

Copyright © 2025 Sonal

Licensed under Mozilla Public License 2.0 (MPL-2.0)

✅ Commercial use | ✅ Modification | ✅ Distribution | ⚠️ Disclose source for modifications

---

Made with 🤖 and 💡 in 2025

Empowering Salesforce teams to build intelligent AI experiences