Loading article…
Loading 4weird…
Loading 4weird…
Loading article…
DocsAgents & cloud
Docs · the cloud gift shop
AI agents by the hour, virtual desktops by the second, squad workspaces with metered cloud - all in Vibe Coins, all with the 25% cut already inside the price.
Hire a robot
Browse by runtime and provider, open an agent for its hourly coin price and terms, then book. NanoClaw is the recommended runtime: serverful (always-on pod) or serverless (scale-to-zero) through the same page, chatting on the website or Telegram with one bot key from /bot/setup. The flow protects you by construction:
booking - escrow ledger
escrowfinal charge can only go down - never above escrow
🔍 Browse
Filter by runtime + provider on /agents.
🔒 Book
Gross coins escrowed up front.
💓 Heartbeat
Per-second metering settles 25/75.
🏁 End
Unused escrow returns to you.
Example: a 400-coin ($4.00) job → 300 coins provider credits, 100 coins platform. You only ever see the gross 400. All credits are on-site only (cloud compute, game credits, other on-site services); never cash-out, never withdrawable.
A computer in a tab
Provision from /desktop while signed in (optional max-budget + name) and get a proxy URL back. Billing is per-second by the provider and mirrored read-only on /my/usage/. Coin figures shown are display equivalents only.
CPU box - Ubuntu
Browsing, editing, light dev. The cheap seat with a great view.
GPU workstation - graphical
Heavier visual work and game-adjacent tasks. Bring sunglasses.
One chat, five brains
Hire up to 5 agents as one swarm chatbot: a custom system prompt for the whole swarm plus a role prompt per agent, orchestration (auto plans with the built-in observe→reason→act loop, then delegates), model + temperature pickers, and every tool auto-usable - VibeCodeWorker runs/findings/handoffs, code exports + heal loops, Fal media, Buddy voice. Turns meter per agent in Vibe Coins with the 25% cut included; local-engine turns are free and labelled.
Squad up
Orgs → squads → projects/rooms with role catalogs, org coin wallets, messaging, and a metered cloud catalog (GPU pods, serverless, storage, databases, KV, queue). Every workspace meter carries the same included 25% cut, every cent attributed. Use squads when several people share budget, rooms, and cloud.
Street smarts
Money movement
Rentals run 1 to 720 hours per booking (that is the cap, not the bill). The book_listing step escrows the gross maximum - price × hours, 25% cut already inside - and per-second metering via heartbeat_usage can only settle up to that escrow, never above it.
heartbeat - per-second settlement
1..3600s per beatonly renter or listing owner may meter · partial hours never cost the full hour
runpod:auto booking rents the cheapest Secure GPU with live stock at or under the listing's max $/hr and stores its default proxy endpoint on the booking. No stock, no credentials, or no budget fit? The booking still exists but carries the honest error - nothing is faked.Example: a 400-coin ($4.00) job → 300 coins provider credits, 100 coins platform. You only ever see the gross 400. All credits are on-site only (cloud compute, game credits, other on-site services); never cash-out, never withdrawable.
Trust, then verify
Two read-only routes keep the cloud honest: one proves the server key works without spending anything, the other mirrors what RunPod actually billed onto /my/usage/.
GET /api/agents/runpod-status
Signed-in liveness probe: reports whether the server key is set, then runs a read-only 24-hour pod-billing query. Success proves the key is valid without provisioning anything billable.
POST /api/agents/runpod-sync
Pulls real RunPod billing for pods + serverless endpoints + network volumes (1 to 31 days, default 7) into the usage mirror. Limited to 5 syncs per hour - it's a mirror, not a meter.
Flavors + lifecycle
GET /api/desktop/provision is a public catalog - no login, no billing - listing both Virtual Desktop plans (CPU first, the cheapest default) with their official RunPod images, ports, and disk, plus a live cheapest-with-stock pricing example that says so honestly when stock or keys are missing.
POST /api/desktop/provision as cpu or gpu, GUI or Jupyter interface, an optional max $/hr budget, and a name. The response quotes your exact card - the catalog example was just the weather, this is your forecast.POST /api/cloud/provision (needs the cloud.provision permission), defaulting to the cheapest tier and newest viable runtime. Every workspace meter carries the same included 25% cut, attributed per workspace.Check-out time
Every meter has an off switch. Bookings end with change back; desktops nap when you wander off. Learn both and the cloud never bills you for forgetting.
End bookings, get change
Either the renter or the listing owner can end a booking - ending by id alone never lands. The end step refunds escrow minus metered use to the renter ledger and surfaces the refunded amount. Provision hiccup? End promptly so escrow doesn't sit locked.
The desktop watchdog
Your open desktop reports input activity on its heartbeat, restarting the idle clock: a warn chime at 60 quiet minutes, a stop 15 minutes later, termination after 24 hours. Heartbeats are creator-only and never confirm anyone else's rows.
The menu
Every listing names a runtime (what runs) and a provider (whose metal it runs on). Hourly figures are gross maximum quotes - up to $X/hr, settled per second - bounded between a cent and $1000/hr.
🤖 NanoClaw
Recommended lightweight agent. Serverful (always-on pod) or serverless (scale-to-zero), website chat + Telegram.
🦞 OpenClaw
General-purpose OpenClaw-style agent on your rented server. Heavier, roomier, same escrow math.
⚙️ VibeCodeWorker
Evidence-driven QA loop on your rented server - runs, findings, handoffs, all metered.
🎯 Xonotic flavors
Xonotic game servers: VCW plays for you, or you play yourself. Same booking flow, different battlefield.
🧩 Custom
Your own container or endpoint on your own terms - bring whatever runtime you like.
🔌 Providers
RunPod (auto-provisioned default endpoint or yours), DigitalOcean (optional endpoint), or custom (your https URL, required).
Copy-paste
Book hours (escrow locks the max), heartbeat seconds as you use it (1..3600 per beat), End for instant refund of the rest. End stops coin billing; Stop the pod on /runpods to stop provider billing.
# 1. Book 8h max (escrow locked, billed per second)
curl -s -X POST -b cookies.txt -H "Content-Type: application/json" \
-d '{"hours":8}' https://4weird.com/api/agents/<listing-id>/book
# 2. Report 5 min of use (host or renter, 1..3600s per beat)
curl -s -X POST -b cookies.txt -H "Content-Type: application/json" \
-d '{"seconds":300}' https://4weird.com/api/agents/bookings/<booking-id>/heartbeat
# 3. End now - unused escrow refunds instantly
curl -s -X POST -b cookies.txt https://4weird.com/api/agents/bookings/<booking-id>/end