Workflow

A Workflow is how you configure a Content Engine: a visual graph of nodes connected by edges. Data enters at one end, flows through generation and quality steps, and leaves as delivered content. It is the single place where everything that happens to your data is defined.

Entry Point → (transform) → generate → quality gate → route → Deliver

How data flows

1. Entry Point receives each data object you upload.
2. Nodes transform, enrich, and generate content from it.
3. Routing nodes split the flow — by language, segment, or test variant — so one object can produce many outputs.
4. Quality gates pass or hold content based on score.
5. Deliver nodes send the finished content to your systems.

Each path through the graph represents one combination of language, segment, and variant — so one object can yield many content entries.

Node types

Node	What it does
Entry Point	A named entry point the Data API uploads to. (Data can't be written to a pool directly.)
Data Pool	Store incoming objects in a data pool, or read related data from one to enrich an object.
Custom Logic	Run JavaScript to clean, reshape, enrich, or route data.
AI Writer	Generate content with a foundational model, driven by a Declaration. Stochastic.
Deterministic Writer	Generate content from a Ruleset. Reproducible.
AI Rewriter	Rewrite existing copy with a model.
Existing Content	Bring existing text (a data field) into the pipeline as content with a content id — e.g. an existing product description.
Compose	Combine several generated parts into one piece of content.
Guardrail Gate	Pass or hold content based on its guardrail score.
QA Gate	Hold content that misses your QA thresholds (score, max errors / warnings). Placed after generation.
Language Router	Split the flow per language (with an "other" fallback).
Segment Router	Split the flow per segment.
Variant Splitter	Split into named variants (e.g. A/B) — each becomes its own content entry.
Batch	Gather the results of a bulk run into one array, for delivery.
Deliver (Webhook)	Send content to a webhook, optionally HMAC-signed.
Deliver (HTTP)	Send content via a custom HTTP request.

Guardrails & quality

Generated content is scored against guardrails — your rules for brand voice, formatting, data consistency, and more. A Guardrail Gate or QA Gate in the workflow then decides whether a result passes automatically, is held for review, or is rejected. This is how axite lets you supervise the system instead of editing every draft.

Branches

A workflow can have multiple branches (like main and an experiment branch), each with its own graph. You build and test changes on a branch, then publish it — so you can iterate on generation without disrupting what's live.

Uploads always use the published workflow, so a new Entry Point becomes available only after you publish.

Most engines are simple

A first workflow is often just: Entry Point → AI Writer → Guardrail Gate → Deliver. Add routing, rulesets, or custom logic only when a use case needs it.