AI features
Drafting, translation, summaries, autofill, tagging, alt text, audits, search and the agents that enrich and moderate content, all validated like a human edit.
Every AI feature in contentworker follows one rule: an AI can never produce content a person couldn’t. Generated values are checked against the content model with the same validation as a human edit, and most features return a proposal you look at before anything is saved.
In the entry editor
Section titled “In the entry editor”| Feature | What it does | API (POST …/entries/…) |
|---|---|---|
| Generate | Drafts a new entry of a type from a prompt | …/generate |
| Canvas | Turns free-form prose into a structured entry, field by field | …/canvas |
| Translate | Translates an entry’s localized fields into another locale | …/{id}/translate |
| Summarize | Writes a summary, optionally into a field | …/{id}/summarize |
| Autofill | Fills one empty field from the rest of the entry | …/{id}/autofill |
| Suggest tags | Proposes tags from your taxonomy, and new ones | …/{id}/suggest-tags |
| Audit | Reviews an entry for problems, and can open tasks for them | …/{id}/audit |
| Moderate | Classifies an entry’s text for policy problems | …/{id}/moderate |
The editing features take "apply": true to save the result as a new draft version straight
away; without it, they return the proposal for you to accept. Either way the result is a draft:
nothing AI-generated is published without someone (or a workflow) publishing it.
On assets, Alt text describes an image for accessibility, and Auto-tag proposes tags.
AI actions
Section titled “AI actions”An AI action is a saved prompt your team reuses: “rewrite for a younger audience”, “write a meta description”. It has a prompt template with variables, an optional target field, and runs against an entry like the built-in features.
Search
Section titled “Search”Delivery search is hybrid by default: it combines keyword matching with semantic similarity,
so a search for “running shoes” also finds “trainers”. Choose mode=semantic or mode=lexical
to use one or the other. The editor also finds related entries and near-duplicates by
meaning. See Reading content.
Agents
Section titled “Agents”Agents run on their own, on publish or on a schedule:
- Enrich fills empty text fields on a newly published entry.
- Moderate classifies the entry and records a hold when something is flagged, or when the model gives no usable answer.
- Curate proposes improvements to text that’s already there.
- Repurpose derives variants for other channels, such as a social post or a newsletter blurb.
Each run is recorded in the agent run log with its decisions and token usage. Whether an agent’s changes apply directly or wait as an agent review is a setting; reviewing first is the default posture, and repurposing always waits for review.
Model tiers
Section titled “Model tiers”Features ask for a tier, not a specific model: fast for classification and tagging,
balanced for drafting (the default), flagship for the hardest writing. Models run on
Cloudflare Workers AI, and your workspace maps each tier to a model, so a model can change without
changing a workflow or an integration.
Usage is metered against your workspace’s AI budget.