Most integration tools I’ve used over the years were impressive on the surface—rich UIs, hundreds of connectors, and quick demos.
But once things moved closer to production, the problems started showing up.
The problem I kept running into
Tools like workflow automators and enterprise integration platforms usually break down in the same ways:
Execution-based pricing that grows unpredictably
Queues, retries, and error handling hidden behind abstractions
Self-hosting treated as a secondary option
Too much emphasis on UI, not enough on runtime behavior
These tools work well for demos.
They become painful when reliability and control actually matter.
What I actually needed
Instead of more UI, I wanted:
A runtime-first integration engine
Explicit control over queues, retries, and parallelism
Predictable costs by running on my own hardware
A simple way to define flows without locking myself into a SaaS
That’s what led me to build Spubhi.
What Spubhi is (and isn’t)
Spubhi is a self-hosted integration runtime, built with a few clear constraints.
What it does:
Executes JSON-defined flows
Supports API-triggered and queue-triggered executions
Uses a payload-based transformation DSL
Runs on Docker / VM / server
Focuses on reliability over visual design
What it doesn’t try to be:
Not a Zapier replacement
Not a no-code tool
Not a marketplace of connectors
It’s built for developers who want control, ownership, and transparency.
Lessons learned while building it
A few things surprised me while working on this:
Queues are harder than UI — but also more valuable
Retry logic becomes complex very quickly in real systems
“Easy” integrations hide painful failure modes
Simplicity at runtime matters more than feature count
Most of the real work wasn’t writing code—it was deciding what not to abstract away.
Why I’m writing this
I’m not launching anything here.
I’m sharing this because I want feedback from people who’ve dealt with integration systems in production.
If you’ve used tools like MuleSoft, n8n, WSO2, or homegrown scripts:
What part frustrated you the most?
Where did things break in real usage?
What would instantly make you reject a new integration tool?
I’m building this in the open and learning as I go.
Thanks for reading.
Top comments (0)