Run Your AI AgentsAt The Lowest Cost.

Without the repeats.

Open source. Open weights. Open proof.

the terminal

Your Terminal. Our Machine.

Claude Code, unchanged. Codex, unchanged. Every line they never had to read, counted.

Cuna CLI installs as @cuna_labs/cli.
Cuna CLI

Cloud

~/acme-apicuna opencode
OpenCode
ship the refund flow. never touch the live keys.
read src/payments/refund.ts
edit src/payments/refund.ts
12 additions · 3 removals · keys untouched
the price

The Same Work. A Smaller Bill.

Pick the model, see its rate, and pay for fewer tokens than you sent.

A 1457-token run written at about 360 tokens per second on gpt-oss-120b, then repeated.

~360tok/s

input
$0.15/M
cached in
output
$0.60/M

USD per million

the context

Change the Provider. Keep the Context.

Today the machine keeps your files, tools and state through every pause. Keeping them across providers is next.

the record

Every Run On The Record

Cuna runs and records the work today. Ring0 is the intended shared-proof layer.

Ring0THE RULESTHE KEYSTHE RECORDTHE BUDGETBLOCKCHAIN L1 · POST QUANTUM

The work ends. The proof remains.

Grant a mandate, put an agent to work, keep the proof. The keys never leave your hands.