Durable execution, hand-rolled one fragile branch at a time.
Teams handling real inbound traffic — WhatsApp orders, email RFQs, inbound leads — hit the same wall. Every client wants a slightly different flow, so the work turns into one more branch in a growing pile of custom code. The flows need to wait, retry cleanly, and never double-charge or double-send.
GridFlow takes the proven durable-workflow shape and aims it squarely at AI-driven B2B operations.
Visual builders don't ship durability guarantees. Serious workflow engines do — but they're general-purpose infrastructure you operate yourself, and none are built around AI and messaging channels.
An engine for inbound, not another builder.
Understands inbound work
First-class AI blocks extract structured data from a messy email, classify intent, and generate a reply — with the cost of every run tracked as real data, not buried in logs.
Talks on your channels
WhatsApp, email (Gmail and Microsoft), and signed webhooks at launch — with new channels added as adapters, not rewrites.
Keeps your domain yours
GridFlow never becomes your system of record. Catalog, inventory, pricing and CRM stay in your systems, exposed as REST endpoints under a signed contract. The engine owns the workflow; you own the business.
Every run is durable, suspendable, and replayable.
This is one real flow shape. Press play, or drag the scrubber to move through the event log — forward and back. Retries are safe to repeat; the flow pauses for a human, then resumes exactly where it left off.
B2B distribution, wholesale, and Inventory-led businesses.
If your ops team is drowning in inbound that's almost structured, and your engineers are tired of writing a new orchestration path per client — that's the fit.
- WhatsApp as a primary sales channel
- Email RFQs quoted by hand today
- A CRM fed by messy, inconsistent inbound leads
- General-purpose "agent platforms"
- Visual-builder-first teams
- Real-time sub-100ms use cases
- Buyers who require SOC 2 / on-prem before they start
We're honest about scope.
Three flows it runs today.
GridFlow's first tenant is SourceGrid, a live WhatsApp commerce platform for automotive parts. Three flow patterns run through the engine — same durability guarantees, three very different shapes.
WhatsApp commerce
A multi-turn conversation that walks a buyer from "need brake pads for a Swift" through cart, checkout, address, and payment. It tolerates editing, back-navigation, and abandonment — modeled as a durable state machine, not a script.
Email RFQ → quote
Read the buyer’s email, match parts against the catalog, check inventory, and reply with a priced quote. When the AI isn’t confident, the flow pauses for a human to review before anything goes out.
Email → CRM
Capture an inbound deal, de-duplicate it, and push it into HubSpot, Zoho, or Salesforce. Fire-and-forget, no conversation needed.
We don't oversell roadmap as reality.
- The three flows above, in production for SourceGrid
- WhatsApp + email + webhook channels
- The AI, logic, and HTTP block library
- The visual editor and run inspector
- The documented host contract
- Parallel / DAG steps
- Self-serve onboarding
- Billing integration
- More channel adapters (Slack / SMS)
- Formal SOC 2 certification
On the post-launch roadmap — not pretended-present.
Anyone can write a workflow loop in a weekend.
What's hard to copy is everything around it.
AI blocks tuned on real B2B traffic
Schema-guided extraction trained against actual RFQ emails and parts enquiries, with the prompts and evals living with us.
Channel adapters that handle the real world
WhatsApp's 24-hour messaging rules and template categories, Gmail and Microsoft Graph push (not brittle IMAP polling), baked in.
Operational multi-tenancy
Per-tenant limits, a real secrets vault, envelope encryption, scoped observability — the unglamorous infrastructure that takes months to get right.
Domain depth
The engine knows what auto-parts and wholesale flows actually look like. Generic agent platforms don’t.
An honest look at the alternatives.
We're not ✓ on every row — parallel/DAG steps are a deliberate v1 gap. Here's where GridFlow fits.
| Capability | GridFlow | Hand-rolled | Temporal | Visual builders |
|---|---|---|---|---|
| Durable retries & suspend/resume | ||||
| Replayable event history | ||||
| AI extraction & intent blocks | DIY | |||
| WhatsApp / email channels built-in | ||||
| Nothing to operate (managed, REST-first) | ||||
| Versioned flows, no diagram-spaghetti | DIY | |||
| You keep the system of record | ||||
| Parallel / DAG steps | soon | DIY |
See GridFlow run your flow.
If you're hand-coding orchestration per client, or standing up Temporal just to get suspend-and-resume, there's a faster path.