Tealfabric

ProcessFlow execution

Trigger → integrate → LLM → push → audit

01 Trigger02 Integrate03 LLM04 Push05 Audit

Process: vendor_intake_enrichment · execution_id: pf_8f2a

AI agent chat

sync · agent trigger

“Enrich vendor Acme Parts — pull CRM account and classify industry.”

WebApp form

sync · form submit

{ "vendor_name": "Acme Parts", "tax_id": "FI-12345", "contact": "ops@acme.fi" }

Parent process

async · process-to-process

onboarding_pipeline → queue subflow · cursor: batch 12/40

step: receive_input → merged process_data.vendor

Simulated flow — governed step code from AI chat, WebApp, or parent process through integrations and LLM to downstream systems

Governed automation

“Write real code. Run inside guardrails—prototypes become production workflows without a rewrite.”

Platform

Process Automation — step-based workflows in a safe sandbox

Model work as a sequence of steps with PHP, JavaScript, or TypeScript. Trigger from APIs, WebApps, schedules, and agents—then call integrations, DataPool, LLM, and files inside tenant guardrails.

Book a demoVendor intake workflow

ProcessFlow is how work moves through Tealfabric—not a separate scripting tool beside the platform.

Teams usually choose between no-code builders that are hard to extend, custom scripts that are fragile and insecure, or enterprise BPM that is heavy and disconnected from data and integrations.

ProcessFlow offers a third path: step-based automation with real code, inside a sandbox that enforces safety. Every step runs in a controlled runtime where code can use platform services but cannot bypass tenant boundaries or platform security.

  • Step-based automation with real code—inside a sandbox that enforces safety
  • Not a separate scripting tool: how work moves through schedules, WebApps, APIs, and DataPool
  • PHP, JavaScript, and TypeScript per step—with platform services built in
  • Sequential and parallel strategies; sync for speed, async for scale
  • Trigger from API, WebApp, cron, events, webhooks, or another process
  • Integrations, DataPool, files, keystore, OCR, and LLM from step code—no credentials in snippets
  • Process-to-process composition, batch continuations, and structured step errors
  • Execution history, step logs, and queue visibility alongside platform monitoring

ProcessFlow decides what to do. Integrations do the talking.

Production patterns

Operational automation teams ship without standing up separate job runners.

  1. webhook → OCR
    → CRM · DataPool

    Invoice processing

    Webhook receives a file → OCR extracts text → validation step → CRM integration updates the account → result written to DataPool.

  2. event trigger
    parallel checks

    Customer onboarding

    Event trigger on new user → parallel checks (identity, compliance, provisioning) → notification step → WebApp shows live status.

  3. cron → ERP
    TypeScript transform

    Nightly operations

    Cron trigger → pull from ERP integration → transform in TypeScript → batch write to DataPool → summary email to ops.

  4. WebApp sync
    async queue

    Self-service portal

    WebApp submits form → sync process validates → async process handles heavy work → user polls queue status.

  5. PDF OCR
    LLM classify

    Document intelligence

    Upload to tenant storage → PDF OCR step → LLM summarization → route by classification.

End-to-end flow

Agent chat, WebApp, or parent process triggers a governed multi-step run.

A vendor intake process receives input from an AI agent, a self-service form, and an upstream onboarding pipeline. Steps call Salesforce, run LLM classification, push to HubSpot, write to DataPool—and leave step logs and queue history for operators.

ProcessFlow execution

Trigger → integrate → LLM → push → audit

01 Trigger02 Integrate03 LLM04 Push05 Audit

Process: vendor_intake_enrichment · execution_id: pf_8f2a

AI agent chat

sync · agent trigger

“Enrich vendor Acme Parts — pull CRM account and classify industry.”

WebApp form

sync · form submit

{ "vendor_name": "Acme Parts", "tax_id": "FI-12345", "contact": "ops@acme.fi" }

Parent process

async · process-to-process

onboarding_pipeline → queue subflow · cursor: batch 12/40

step: receive_input → merged process_data.vendor

Simulated flow — governed step code from AI chat, WebApp, or parent process through integrations and LLM to downstream systems

Capabilities

Composable steps, flexible timing, and operable execution—not no-code only or scripts on a server.

Design in the platform library, test before go-live, and run from anywhere your business needs work to start.

  1. steps · conditions
    merge semantics

    Step-based process modeling

    Multi-step workflows with conditions, branching, and merge semantics.

    Model work as a sequence of steps—each runs logic that validates data, calls a system, transforms a file, or decides what happens next. Output from each successful step merges into shared state for downstream steps.

  2. PHP · JS · TS
    per step

    PHP, JavaScript, and TypeScript

    Mix languages per step—meet teams where they already work.

    PHP for legacy snippets and rapid procedural logic; JavaScript for JSON-centric transforms; TypeScript for typed step logic and safer refactors. The platform detects or respects the language you choose per step.

  3. sequential · parallel
    sync · async

    Sequential, parallel, sync, and async

    Strict order when steps depend on each other; parallel when branches are independent.

    Run now and wait for an answer, or queue and check later. Same process definition—choose timing per invocation. Suited to approvals, enrichment chains, and overnight imports alike.

  4. API · WebApp
    cron · webhook

    Trigger from anywhere

    API, WebApps, schedules, platform events, webhooks, and process-to-process.

    External systems trigger with tenant API credentials. User-facing apps call processes synchronously or in the background. Cron, events, and webhooks start runs automatically—one workflow can queue another without blocking.

  5. guardrails
    tenant-scoped

    Sandbox guardrails

    Real code inside a governed runtime—not raw server access.

    Step code runs in an isolated environment with approved platform services. Shell execution, raw network calls, direct database drivers, and environment tampering are blocked at save time and runtime.

  6. integrations
    DataPool · LLM

    Platform services built in

    Integrations, DataPool, files, keystore, OCR, LLM, and notifications from step code.

    Every step starts with injected context and service handles—tenant scope, process input, configured integrations, internal API, database, DataPool, files, secrets, logging, and document intelligence.

  7. process-to-process
    batch cursor

    Process composition and batch work

    Fan-out subflows, event chains, and cursor-based continuations.

    Processes trigger other processes for reuse and event-driven chains. Iterate over bounded lists in one step, or hand off the next batch by queuing another run with a cursor—large datasets without overloading a single execution.

  8. step logs
    queue · Monitor

    Built to be operated

    Step logs, execution history, queue visibility, and structured error codes.

    When something fails at 2 a.m., teams see which step, which run, and why—not a black box. Process activity sits alongside integrations, security, and data audit in Monitor.

Sandbox services at runtime

Authors write normal step logic. The sandbox supplies context, services, and guardrails so snippets stay small and focused. Configurable capability profiles let operators tighten or broaden what a run may do.

Process input & context

Read trigger payload and accumulated step output

Tenant scope

Every operation stays inside the current tenant

Integrations

Call configured connections to external systems

Internal API

Reach platform APIs with execution authorization

Tenant database

Query and update transactional data safely

DataPool

Insert and update governed datasets from steps

Files

Read and write tenant-scoped storage

Keystore

Retrieve secrets at runtime—never hardcoded in snippets

Logging

Write execution logs for debugging and audit

Document intelligence

PDF and image OCR from tenant files

LLM assistance

AI-powered steps within platform policy

Notifications

Email and messaging from workflow logic

Languages per step

PHP

Existing customers, legacy snippets, rapid procedural logic

JavaScript

Modern step code, JSON-centric transforms, async service calls

TypeScript

Typed step logic, larger teams, safer refactors

Pairs with Integrations

ProcessFlow is the orchestration layer; Integrations are the connection layer. Configure Salesforce, SFTP, or email once—call from any step without credentials in code. Combine multiple systems in one governed workflow and land outputs in DataPool for reporting and follow-on processes.

Integrations & Connectors

Turn business rules into production automations

See ProcessFlow with sandbox guardrails, triggers, and integration steps.

We walk through step design, sync and async execution, keystore secrets, LLM steps, DataPool writes, and execution history—on the same runtime as integrations and governance.

Book a demoStart a pilotIntegrations