FlowMason

Pricing

Built for Salesforce teams.
Priced for AppExchange.

You bring your own LLM API keys. FlowMason never charges per token — your AI costs go directly to your provider.

Pricing finalised at AppExchange launch — get early access now

Starter

For individual developers and small teams exploring AI pipelines

  • Up to 3 active pipelines
  • 1 LLM provider
  • Flow Builder invocation (@InvocableMethod)
  • Apex facade (PipelineRunner.run())
  • Basic audit log (30-day retention)
  • FMTestMocks test framework
  • Community support
Get Early Access
Most Popular

Professional

For Salesforce teams shipping AI features in production

  • Unlimited active pipelines
  • All 7 LLM providers + provider switching
  • Pipeline Studio (visual builder + SFDX export)
  • Full audit log + cost reports (unlimited retention)
  • All 6 LWC drop-in components
  • REST API endpoints
  • Trigger framework + Platform Events + Scheduler
  • Async execution + GovernorMonitor
  • Email support + response SLA
Get Early Access

Enterprise

For orgs with compliance, multi-org, or private AI requirements

  • Everything in Professional
  • Multi-org deployment
  • Private/on-premise LLM support (Ollama, custom Bedrock)
  • Custom permission set configuration
  • SSO integration support
  • Dedicated onboarding + architecture review
  • Custom SLA + named support contact
Contact Us

Full feature comparison

Feature Starter Professional Enterprise
Active pipelines 3 Unlimited Unlimited
LLM providers 1 All 7 All 7 + custom
Flow Builder (@InvocableMethod) ✓ ✓ ✓
Apex facade (PipelineRunner) ✓ ✓ ✓
FMTestMocks (unit testing) ✓ ✓ ✓
Pipeline Studio + SFDX export — ✓ ✓
LWC drop-in components (6) — ✓ ✓
REST API endpoints — ✓ ✓
Trigger framework + Platform Events — ✓ ✓
Async execution + GovernorMonitor — ✓ ✓
Audit log 30 days Unlimited Unlimited
Cost reports — ✓ ✓
Private / on-premise LLM — — ✓
Multi-org deployment — — ✓
SSO integration — — ✓
Support Community Email + SLA Named contact
Cost model

Beyond the FlowMason subscription

FlowMason itself is per-org subscription. LLM calls bill to your vendor account directly; we never mark up. Below is how to forecast the variable line.

Direct mode (vendor APIs)

Calls hit Anthropic / OpenAI / Bedrock / etc. Vendor bills you directly via Named Credential. No Einstein Requests consumed.

  • Per-stage Cost__c in Pipeline_Stage_Log__c
  • Aggregated per-run in ExecutionTrace.usage
  • Pricing rows in FM_Provider_Pricing__mdt

Trust-layer mode (Models API)

Calls route through Salesforce Einstein Models API. Bills as Einstein Requests, not USD.

  • Source of truth: Setup → Einstein → Usage
  • BYOLLM Open Connector models 30% cheaper than Salesforce-managed
  • FlowMason logs request counts to FMLog

Tool-calling cost (ADR-013)

Tool-calling adds round-trips per turn but reduces prompt tokens because conversation history lives server-side in FMThreadState, not re-sent each turn.

MetricPre-tool-callingTool-calling on
Round-trips / turn23-5 (capped at orgChatToolCallingMaxCalls, default 5)
Prompt tokensbaseline≥ 30% reduction on long conversations
Wall-clock budgetn/aorgChatToolCallingTimeoutMs default 25s

Net cost depends on conversation length. Short ones slightly more expensive; long ones cheaper.

Order-of-magnitude forecasting

  • Anthropic Sonnet, 1k input / 500 output, $3 / $15 per 1M = ~$0.011 / call
  • OpenAI GPT-4o, 1k input / 500 output, $2.50 / $10 per 1M = ~$0.0075 / call
  • 100 turns/day × 4 round-trips × $0.01 = ~$1.20 / day ≈ $36/month per active chat user

2026-04 vendor rates. FM_Provider_Pricing__mdt is the source of truth.

Cost-optimisation playbook

  1. Cache hot prompts. Set cacheable: true on stages with stable system + user + model + temp. Identical input → zero callout.
  2. Pin smaller models per stage. Classification + extraction usually work with the smallest model in the family.
  3. Drop useFewShot where unused. Few-shot prefix tokens land on every call.
  4. Watch __meta.providerAttempts. High = first provider failing often. Rotate the order in providerFallback.
  5. Tune orgChatToolCallingMaxCalls. Over-permissive lets the model loop; under-permissive aborts useful flows.
Read the tool-calling docs →

Not sure which tier fits?

We'll tell you honestly. Most teams start with Professional.

Talk to us