Live · Shared cache for AI agents

Crack the page once. Share the kernel.

Aicorn is a shared, credit-metered cache for agent-friendly web content. One agent extracts. The rest of the forest harvests. Contributors earn credits when others read what they cracked.

Aicorn mascot
20×
Cheaper per cached read
400
Tokens on a cache HIT
8,000
Tokens on a cold MISS
5,000
Free credits at signup

The forest beats the lone tree.

Every AI agent re-extracts the same HTML from the same websites and burns the same tokens cleaning it up. Aicorn is the squirrel-network that ends the duplication.

01

Agent requests a URL

Any agent — Claude, GPT, Perplexity, your own — points its fetch at the Aicorn proxy.

02

First fetch cracks the shell

Cache miss. Workers AI cleans the page into agent-ready markdown. The fetcher pays extraction + 10 credits and becomes the URL's contributor.

03

Every other agent harvests

Cache hit. Reader pays 10 credits. The original contributor earns 9. Network pockets 1 credit of margin. Forever sybil-proof.

The credit ledger

Contribution = future consumption rights.

Same shape as BitTorrent ratios and Filecoin storage proofs — but for agent-readable web content. Anyone who pays to extract a page is buying future free reads for themselves and the network.

ledger.rules
  • Reading any cached pagecosts 10 credits
  • First fetcher of a URLpays extraction + 10, becomes contributor
  • Each subsequent readearns the contributor 9 credits
  • New usersget a 100-credit signup grant
  • Sybil mathself-dealing always loses 1 credit per round
The visceral moment

Two agents. Same URL. Ten seconds apart.

Agent AX-Cache: MISS
Tokens
8,000
Cost
$0.02

Cracks the shell. Becomes contributor.

Agent BX-Cache: HIT
Tokens
400
Cost
$0.00

Pays 10 credits. Agent A earns 9.

Same content. Same agent quality. 20× cheaper.

All Cloudflare

One Worker. One deploy. The whole network.

Agent request
   ↓
Cloudflare Worker  (header detection, routing, ledger)
   ↓
KV cache lookup: is this URL extracted?
   ├── HIT  → KV read markdown
   │         → charge reader 10 credits
   │         → credit contributor 9 credits
   │         → return markdown  [X-Cache: HIT]
   │
   └── MISS → fetch origin HTML
              → Workers AI extraction
              → write to KV
              → charge fetcher (extraction_cost + 10)
              → register fetcher as contributor
              → return markdown  [X-Cache: MISS]

Cloudflare Workers

Proxy, ledger routes, dashboard — one TypeScript project, one deploy.

Workers AI

Llama-class extraction model turns raw HTML into clean markdown at the edge.

KV

Cache index, extracted bodies, ledger balances, live stats counters.

No external services

No databases to provision. No queues. Cloudflare end-to-end.

Try it now

Three steps to a cheaper agent.

No infrastructure to set up. No model to host. Just sign up, install, and ship.

01

Create your account

Sign up in under a minute and instantly receive 5,000 free credits — enough to harvest thousands of cached pages.

Create free account
02

Install SKILL.md on your agent

Drop our SKILL.md into Claude, Cursor, your custom agent, or any tool-using LLM. It teaches your agent to read the web through Aicorn.

Get SKILL.md
03

Start using and saving tokens

Every cached page costs a fraction of a fresh extraction. Watch your token bill drop from the first request — and earn credits when others read what you cracked.

See the economics

Stop paying to clean the same page twice.

Create a free account, drop SKILL.md into your agent, and start saving tokens on your very next request.