FLOW MASON
Coming Soon - Beta Available

Flowmason for Salesforce

AI Pipeline Orchestration Native to Your Org

Build complex AI workflows that run entirely within Salesforce. Same pipeline JSON works locally and in production. No external infrastructure required.

View Live Demos Read Documentation

Live Demo Results

9 pipelines tested on December 14, 2025

8/9
Pipelines Succeeded
17
LLM Calls Made
~$0.11
Total Cost
<20%
Governor Limits Used

Local-First Development

The same pipeline JSON works identically on your machine and in Salesforce. Test locally in seconds, deploy when ready.

1

Develop Locally

$ fm run pipeline.json \
--input data.json
# Instant feedback
# Full debugging
# No deployment wait
2

Deploy JSON

Same JSON, no changes needed
Store in Static Resource
Or Custom Metadata
Version control friendly
3

Run in Production

// Apex
PipelineRunner.execute(
pipelineJson,
input
);
// Triggers, LWC, Flows

What Was Hard Before

Flowmason eliminates the complexity of building AI-powered workflows in Salesforce

Before: AI Integration

Custom HTTP callouts, JSON parsing, error handling, retry logic, token counting, cost tracking... Each LLM call requires 50+ lines of Apex.

With Flowmason

Single generator component. Built-in retries, token tracking, cost estimation.

Before: Complex Routing

Nested IF/ELSE statements in Apex. Hard to visualize, harder to maintain. Business logic buried in code.

With Flowmason

Visual router + conditional components. Logic is in JSON, not code.

Before: Batch Processing

Custom iterators, manual governor limit tracking, complex state management across Queueable chains.

With Flowmason

foreach with collect_results. Automatic result aggregation.

Before: Error Recovery

Try/catch blocks everywhere. Manual fallback logic. Inconsistent error handling across different code paths.

With Flowmason

trycatch component with automatic fallback paths. Consistent, declarative error handling.

Before: Testing AI Pipelines

Deploy to scratch org, wait 30+ seconds, test, fix, repeat. Each iteration costs real API calls and time.

With Flowmason

Test locally with fm run. Instant feedback, full debugging. Deploy working code.

Live Demo Results

9 production-ready pipelines tested with real AI (Claude 3.5 Sonnet)

AI
Success

Customer Support Triage

Classify tickets and generate responses

Time
9.1s
LLM Calls
2
Tokens
~600
Cost
~$0.005
Non-AI
Success

Data Validation ETL

Schema validate, transform, filter

Time
83ms
LLM Calls
0
Stages
8
Cost
$0.00
HTTP
Config Required

Multi-API Aggregator

Parallel HTTP + transformation

Requires Remote Site Settings for external API endpoints
AI-Heavy
Success

Content Generation

5 AI outputs from 1 product description

Time
40.7s
LLM Calls
5
Tokens
~3,000
Cost
~$0.02
Non-AI
Success

Error Handling

TryCatch with automatic fallbacks

Time
20ms
LLM Calls
0
Stages
7
Cost
$0.00
Non-AI
Success

Batch Processing

ForEach with result collection

Time
87ms
LLM Calls
0
Stages
7
Cost
$0.00
AI
Success

Conditional Workflow

VIP routing + dynamic branching

Time
3.0s
LLM Calls
1
Stages
14
Cost
~$0.003
AI-Heavy
Success

Book Editor v1.0

Full editorial pipeline with 4 LLM calls

Time
43.8s
LLM Calls
4
Tokens
~4,000
Cost
~$0.03
AI-Heavy
Success

Book Editor v1.1

Streamlined with 5 LLM calls, 3 versions

Time
48.3s
LLM Calls
5
Tokens
6,285
Cost
$0.0445

Real AI Outputs

Actual outputs from the Book Chapter Editor pipeline

Editorial Critique (AI-Generated)

From Book Editor v1.1 pipeline

Strengths

  • - Atmospheric Opening: Cold coffee and rain imagery establishes melancholic mood
  • - Subtle Character Development: Mother-daughter dynamics revealed through dialogue
  • - Authentic Dialogue: Natural conversation loaded with subtext
  • - Effective Physical Details: Mattress dip, flinch, cold coffee ground the scene

Areas for Improvement

  • - Missing apostrophes in contractions
  • - Relationship to deceased unclear (intentional mystery or oversight?)
  • - Consider adding more emotional specificity
Grade: B+ - Strong atmospheric writing needing minor polish

Production Ready

All pipelines run well within Salesforce governor limits

17%
Callouts Used
17 of 100 limit
5%
CPU Time Used
~500ms of 10,000ms limit
2.5%
Heap Size Used
~150KB of 6MB limit

Component Library

Pre-built components for common AI and data workflow patterns

AI Components

  • generator LLM text generation
  • critic Content evaluation
  • classifier Categorization

Data Operators

  • json_transform JMESPath
  • filter Array filtering
  • schema_validate JSON Schema

Flow Controls

  • conditional If/else branching
  • router Value routing
  • foreach Iteration
  • trycatch Error handling

Ready to Get Started?

Join the beta program and bring AI pipeline orchestration to your Salesforce org.

Join Beta Program Read Documentation