TPT Conduit

The "ServiceNow Disruptor" — an open, headless, API-first workflow & ITSM engine.

Every Fortune 500 company relies on IT Service Management (ITSM) and workflow engines for everything from IT helpdesk tickets to HR onboarding and asset tracking. That space is dominated by ServiceNow: a proprietary black box that costs millions to license, takes years to implement, and requires certified developers just to change a dropdown. TPT Conduit reimagines ITSM as a modern, programmable, free public good.

Instead of locking users into a rigid UI, Conduit provides a blazing-fast, durable execution engine and treats workflows as code. Developers and IT admins define approval chains, ticket routing, and SLA escalations in simple YAML or TypeScript, version-controlled in Git. A beautiful, white-label React frontend consumes a headless GraphQL/gRPC API.

Releasing TPT Conduit as an Apache 2.0 public good democratizes internal tooling: companies build custom workflows in days, not months, and a community of shared templates erodes the ServiceNow monopoly.

Tech stack

Layer	Choice
Workflow engine	Go, Temporal-inspired durable, fault-tolerant execution
API layer	GraphQL + gRPC for high-performance state queries/mutations
Workflow DSL	YAML and TypeScript, compiled to one shared IR
Frontend	Next.js (React) with white-label theming and a plugin model
Storage	PostgreSQL first; CockroachDB/Cassandra considered for scale

Project layout

/engine      Durable Go workflow engine (domain model, event log, worker, DSL IR)
  /dsl       YAML workflow DSL compiler -> shared IR
  /examples  Reference workflow templates (helpdesk, HR onboarding, asset tracking, approval chain)
/api         GraphQL + gRPC service surface (Phase 2)
/web         Next.js frontend (Phase 4)
/docs        Documentation site (Phase 6)

Architecture principles

Event-sourced & durable. Workflow state is never stored directly. Every effect is an append-only event; run state is always derived by replaying the log. A process can crash at any moment and resume exactly where it left off via Engine.Recover.
Generic / domain-agnostic. A Ticket is just structured Fields; one engine serves helpdesk, HR, and asset tracking alike.
Primitive-based. Approvals, timers, retries, SLA escalation, and routing are composable step kinds, not bespoke code paths.

Quick start

# 1. Start Postgres (applies engine/schema.sql automatically)
docker compose up -d

# 2. Build & test the engine
go build ./...
go test ./...            # in-memory backend
PG_TEST=1 DATABASE_URL='postgres://conduit:conduit@localhost:5432/conduit?sslmode=disable' \
  go test ./engine -run Postgres

Status

Phases 0–1 are implemented: repository scaffolding, the durable Go engine (domain model, event-sourced log + replay, Postgres persistence, in-process worker pool, retry/timer/SLA/approval/routing primitives) with crash-recovery and replay tests. Phases 2+ (API surface, frontend, distributed scale-out) are outlined in TODO.md.

tpt-conduit