Most "AI in your app" turns out to be a chat box bolted onto a product, answering questions about it. We wanted the opposite: an agent that builds the product — that takes a sentence and produces the actual structured artifacts your platform runs on.
WP-PFAgent is that layer, and it's open source (GPL-2.0). You describe what you need in plain language — "a service desk with incidents, an SLA field, and a rule that escalates anything open over 24 hours" — and the agent designs it against documented schemas: entities, fields, forms, business rules, and workflow graphs. Not a mockup. The real definitions the low-code platform and the workflow engine execute.
It writes to the same primitives you would
The reason this works is that the platform underneath is already structured. WP-PFManagement models everything as typed entities, fields, permissions and rules; WP-PFWorkflow models automations as graphs of triggers, branches and functions. Those are documented, machine-legible primitives — so "turn a sentence into an app" is really "turn a sentence into the same objects a human would click together," and then you review and adjust them like anything else.
That means the agent isn't a black box that generates opaque output. It produces entities you can open, rules you can read, workflow nodes you can rewire. If it gets something 90% right, you fix the 10% by hand — you were going to own these definitions anyway.
Bring your own keys, any provider
The agent is bring-your-own-keys. You point it at your own LLM provider — OpenAI, Anthropic, Gemini, DeepSeek, Qwen, whatever you already pay for — and it uses that. Nothing routes through us. It works fine with cheap models; you're generating structured definitions against a schema, not writing novels. And because it's GPL and free, the suite is fully usable without it — the agent is leverage, not a lock-in.
Confirmation before it acts
An agent that can create entities and rules needs guardrails, so it has one: it asks for confirmation before any write or send. You see what it's about to create or change, and you approve it. The agent proposes; you commit. That keeps "designs your schema from a sentence" from turning into "silently rewired your production data model."
Where it runs, plainly
Like the rest of the suite, it's a plugin inside your own WordPress — no external SaaS, no telemetry, your data stays in your database, and the only outbound call is to the LLM provider you configured. The commercial pieces (the low-code platform and the workflow engine the agent designs for) are licensed per domain, refundable, and on sale now — register, buy and self-host today. The agent itself and the Rust executor are open source and free.
There's a worked end-to-end example — a WooCommerce order becoming a ticket, a workflow, an AI triage and a generated RMA file — at project-flash.com/use-case. Docs at /docs.
Describe the app; review what it built; ship it — on your own server, with your own keys.
Happy to answer anything about the schema the agent targets, the confirmation model, or bring-your-own-keys.
Top comments (0)