Push Platform AI
AI at the Edge.
Zero Infrastructure.
A working demo of agentic AI behaviour powered by Hugo and Cloudflare Workers AI. No API keys, no servers, no ongoing compute costs.
01 · Live Demos
Pick a capability. Try it live.
Summarise
Paste any text and get a concise 2-3 sentence summary. Single-shot agent task using Mistral Small 3.1.
Try it →Extract Entities
Pull structured JSON of people, organisations, places, and dates from any text. Typed output from raw inference.
Try it →Classify Sentiment
Get positive, negative, neutral, or mixed classification with confidence scores and reasoning.
Try it →Research Agent
Multi-step agentic loop: plans research questions, answers each, then synthesises a coherent brief. Full trace visible.
Try it →02 · How It Works
The agentic loop. Entirely at the edge.
Static HTML/CSS/JS served via Cloudflare Pages. Fast, cacheable, secure.
JavaScript fetch() sends the task type + input text to a Cloudflare Worker endpoint.
Worker calls env.AI.run() with the selected model. Runs on Cloudflare's global GPU network.
Worker parses model output into typed JSON. Hugo renders the result in the UI.
03 · The Stack
Intentionally minimal. Surprisingly capable.
Hugo
Static site generator. Compiles to HTML/CSS/JS. Deploys to Cloudflare Pages.
Cloudflare Workers
Edge compute. Receives tasks, constructs prompts, orchestrates AI calls, returns JSON.
Workers AI
Model inference at the edge. Mistral Small 3.1 & QwQ-32B. Zero config, no API keys.
Extensible
Add KV (memory), D1 (database), Vectorize (RAG), R2 (storage), Durable Objects (sessions).