A small API.
A clear contract.

Analyze a workflow

POST https://inrigpt.com/api/analyze.php

Requests originate from this website and require a valid, single-use Cloudflare Turnstile token for the demo action. There is no public API key or unrestricted server-to-server endpoint.

{
  "workflow": "proposals",
  "input": "Your sample or redacted source text…",
  "turnstile_token": "A fresh token from the website widget"
}

Supported workflows

• proposals — requirements, evidence, gaps, next steps.
• knowledge — a question answered from supplied procedure text.
• sales — objections, buying signals, and follow-up ideas.
• content — social, newsletter, and video drafts.

Response

{
  "result": "The generated working draft…",
  "workflow": "proposals",
  "model": "@cf/meta/llama-3.1-8b-instruct-fp8",
  "review": "AI draft. Check facts and source evidence before use."
}

Everyday chat and image generation

POST /api/chat.php accepts messages (alternating user/assistant objects with role and content, ending with user) and a fresh Turnstile token with action chat. A new message is limited to 2,000 characters; the recent context is limited to 11 messages and 8,000 UTF-8 bytes. It returns result and model.

POST /api/image.php accepts a prompt of 3?1,000 characters, style (natural, photo, illustration, watercolor, or pixel), and a fresh Turnstile token with action image. It returns a base64 image, mime (image/jpeg), and model. Shared UTC daily limits are 60 chat replies and 10 images, separate from the business demo. Failed inference attempts count. All tools share the Worker limit of three requests per minute per client.

Limits and errors

Input must contain 30–4,000 characters. The demo allows up to three AI requests per minute per client at the Worker, with a site-wide cap of 30 AI requests per UTC day. Failed inference attempts also count toward the cap. Availability is subject to Cloudflare account quotas.

Data and security

AI credentials remain on the server. INRI GPT does not persist demo prompts or outputs. The browser renders generated output as text, and the model has no tools or access to your business systems. See privacy details for the processing flow.