Pricing

Pay for what you build,
not for idle.

A private box from $21/mo. Bring your own model key and pay only for the box — or use our managed models, metered to the token, with smart routing so easy work never pays frontier prices.

Start building →

Just testing? There's a free tier — a real Medium box with 25 active hours/mo and Gemini Flash included. No card.

Free tier · test the basics

$0 /mo — no card
MachineMedium · 4 GB
Active hours25 /mo (hard cap)
ModelGemini Flash + starter tokens
Voicebring your own key

No bill to run up — when the month's hours are used, the box rests until the 1st and your files stay put. Boxes idle for two weeks are recycled (we email you 3 days before). Seats limited.

Start free →

Your box

MachineRAMBring your own key
Small2 GB$21/mo
Medium4 GB$35/mo

One isolated machine that's yours. Includes 250 active hours/mo (~8h/day of the box actually running); heavy always-on use beyond that is billed by the hour — ~$0.08/hr (small) or ~$0.15/hr (medium). Add managed voice or a model at checkout. Cancel anytime.

Managed models · per 1M tokens

ModelInOut
Gemini Flash$1.20$10
Haiku 4.5$4$20
Sonnet 5$12$60
Fugu$16$80
Opus 4.8$20$100
Fugu Ultra$20$120

Optional — skip it entirely with your own OpenAI-compatible key.

Why it's cost-effective

Pay for the model each task needs

Managed model cost spans 12× from cheapest to frontier. Auto grades every request and routes it down the ladder — so a rename never costs what an architecture does.

A typical session — smart-routed vs. always-max

Illustrative · 1M output tokens across a mixed session
Always Ultra
$120
Always Opus
$100
Auto (smart)
$34
~66% lessthan sending every request to a frontier model — automatically.

The Auto ladder

Gemini FlashFuguSonnetOpusFugu Ultra
DifficultyModelBest for
TrivialGemini Flashchat, one-liners
SimpleFugusmall edits, lookups
ModerateSonnet 5multi-file features, debugging
HardOpus 4.8architecture, complex reasoning
ExpertFugu Ultralarge-scale, research-grade
💤

Zero idle cost

Boxes stop after ~15 min idle and wake in seconds, so you only burn hours when you're actually working. Each plan includes 250 active hours/month; only sustained always-on use is billed by the hour beyond that.

📊

Metered to the token

An included monthly allowance, then transparent per-model pricing. No surprise frontier bills.

🔑

Bring your own

Use your own OpenAI-compatible key and pay only for the box — from $21/mo.

Ready when you are.

Spin up your private agent box in minutes — or kick the tires for free first.

Start building →
or start free — no card →