Loom

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.

Loom is named for what the framework actually does. A loom does not create value from a single thread. It turns many separate threads into something structured, durable, and useful. This framework does the same with prompts, tools, context, channels, memory, and governance. The result is not just an LLM call. It is an operational runtime that can hold together in production.

---

🎥 Watch It In Action

Function Agents Demo →

See the framework handle governed AI workflows with intelligent filtering, human-in-the-loop approvals, and error recovery on Salesforce.

---

� Documentation

View Full Documentation →

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

---

💖 Support This Project

Loom 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 public package in force-app contains the core runtime, while the overall framework experience also includes broader packaged capabilities such as additional agent patterns, providers, actions, workflow composition, and UI features.

---

🎯 What is This?

Create AI-powered assistants that can:

• 💬 Chat naturally with users and remember conversation context
• ⚙️ Run focused function-style automations for classification, enrichment, and guided business tasks
• 📧 Process email workflows for triage, draft generation, and routing
• 🔍 Search and retrieve Salesforce data intelligently
• ✏️ Create and update records based on user requests
• 🔄 Execute multi-step workflows with approvals, sequencing, and specialist sub-agents
• 🔒 Respect permissions - agents only access what users can access
• 🎯 Work with multiple AI providers - OpenAI-compatible models, broader provider strategies, and enterprise options

---

💼 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 function-style agents, email workflows, sequential pipelines, and specialist sub-agent patterns for lead qualification, case routing, approvals, and record updates while keeping execution observable and governed inside Salesforce.

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 chat, email, SMS/WhatsApp/webhook, middleware-backed normalized ingress, 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

Public messaging/webhook traffic can first land in a normalized ingress boundary and optional staging object before the core runtime creates InteractionSession__c, InteractionMessage__c, and AgentExecution__c. When staged guest ingress is used, processing is handed off through a Platform Event subscriber configured to run as an internal user instead of continuing in guest context.

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

---

✨ Key Features

Feature	Description
Multiple Runtime Patterns	Conversational and Direct agent runtimes plus a dedicated Pipeline composition subsystem across chat, email, SMS, WhatsApp, API, and sub-agent workflows
Metadata-Driven Capabilities	Define tools, prompts, trust controls, and workflow behavior through Salesforce configuration
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-compatible APIs work out of the box, and the framework supports broader provider strategies for enterprise deployments.
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 with a full execution trail. See exactly what the AI decided and why.

---

🚀 Quick Start

Prerequisites

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

Installation

Option 1: Unlocked Package (Recommended for Quick Start)

Install directly via package URL:

• Sandbox & Scratch Orgs: https://test.salesforce.com/packaging/installPackage.apexp?p0=04tgK0000009qU1QAI

• Production & Developer Edition Orgs: https://login.salesforce.com/packaging/installPackage.apexp?p0=04tgK0000009qU1QAI

After installation:

• Assign permission sets: AIAgentStudioConfigurator (for admins), AIAgentStudioEndUser (for users)
• Configure your LLM provider (OpenAI or any OpenAI-compatible API)
• Create your first agent

Option 2: CumulusCI (Best for Development & Testing)

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 External Client App for API access

Option 3: Salesforce CLI (Source-Based)

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

Optional: Load Test Data

If you need sample data to explore the framework (especially with Option 1 or 3):

1. Deploy the test data factory from the seed-data folder:

   sf project deploy start -d seed-data/main/default -o your-org-alias

2. Execute in Developer Console (or via Anonymous Apex):

   AgentTestDataFactory.createComprehensiveShowcase();

This creates sample agents, capabilities, accounts, contacts, and test scenarios to help you get started quickly.

Configure OpenAI API Key

The framework includes pre-configured OpenAI named credentials. You just need to add your API key:

1. Navigate to Setup → Named Credentials → External Credentials
2. Find and open OpenAIEC
3. Under Principals, click Edit on the principal
4. In the Authentication Parameters section, add:
   • Parameter: OpenAIKey
   • Value: Your OpenAI API key (starts with sk-)
5. Click Save

The OpenAILLM named credential is now ready to use with the framework.

Tip: The framework works well with OpenAI-compatible APIs and can be adapted to broader provider strategies for enterprise deployments. See the Configuration Guide.

Setup

Once your API key is configured:

1. Create or use existing LLM Configuration (references the OpenAILLM named credential)
2. Create AI Agent Definition with identity/instruction prompts
3. Add Capabilities (tools) the agent can use
4. Add Chat Component to a Lightning page or use Quick Actions

👉 Full Getting Started Guide →

---

🏗️ Architecture

Framework Capabilities:

• Conversational strategy for multi-turn chat, email, and external messaging experiences
• Direct strategy for targeted automation and decision support
• Channel-aware routing for chat, email, SMS, WhatsApp, API, and future transports
• Provider-backed webhook/channel seams for external messaging transports such as SMS and WhatsApp
• Sequential pipelines and sub-agent workflows for multi-step orchestration
• Tool execution across data operations, flows, and custom business logic
• Human-in-the-loop approvals, observability, and trust controls
• Flexible support for multiple model-provider strategies

Extension Areas:

• Custom actions for business logic and integrations
• Context providers for domain-specific enrichment
• Additional model providers
• Custom memory strategies
• New execution patterns and agent behaviors

👉 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 © 2026 Sonal

Licensed under Mozilla Public License 2.0 (MPL-2.0)

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

---

Made with 🤖 and 💡 in 2026

Empowering Salesforce teams to build intelligent AI experiences