BabelDag — one engine for business functions in production.
The platform every tai42 application runs on — built on our open-source runtime, delivered as a private tenant your team operates, or self-hosted on your own infrastructure.
What it runs
Not workflow runs — running operations. A BabelDag application is a long-lived machine: it waits, asks a customer or a colleague something, sleeps for a day, wakes on the reply, takes a payment that cannot be faked, and hands the hard cases to a human — holding identity, context, and commitments across restarts and days of silence.
What you get
Deterministic where it matters, generative where it's wanted.
Every application is split in two: explicit flows own the facts, the actions, and the money — prices, lookups, routing, commitments; models own the language — interpreting, phrasing, proposing. The AI has no tools: it cannot fetch data, quote a price, or take an action. Routing between steps is a fixed rule, never the model's choice, and unsafe structure is rejected before it can run. Over time the optimizer watches live behavior and compiles repeated model work into deterministic flows — human-approved, replay-validated, instantly reversible — so the application gets cheaper and more deterministic with use.
Private tenant — or your own infrastructure.
A dedicated, EU-hosted tenant your team operates; or the open-source runtime on your own machines. Your AI keys, any model provider, swappable per step. Your data stays in your tenant or your environment.
Always on our open-source runtime.
The engine underneath is public — read the code, run it, extend it. BabelDag is the hosted platform and enterprise layer built on it — never a different core. (→ /open-source · → Docs)
The visual builder.
Build and change applications as configuration you keep — guardrails in place, every flow versioned, every change reviewable, rollback instant. Swap the model; keep the flows. No vendor holds your business logic, including us.
Audit and observability, by construction.
Every action traced, every flow versioned, every decision attributable — not an add-on, but the way the engine records reality. See what each application actually did: which step ran, which model was asked, which human decided, what moved.
Live in weeks, not quarters.
Applications ship as configuration on the platform, not product code — which is why an engagement is short and what you keep is a running application. Acceptance criteria you verify yourself; a license only if you keep it. (→ /method)
Getting started
Enterprise pricing, sized to your tenant and usage — talk to us. The production readiness report is free: we assess your demo and your integration surface, and you leave with a written report whether or not we work together. No self-serve: the platform ships inside an engagement or a guided onboarding, because production systems deserve a running start.
Looking for something free to try? Our open-source runtime is public (→ /open-source). If you're a builder who serves clients, join the founding waitlist (→ /builders). The platform itself is for teams already in production.