An auth upgrade over managed cloud workspaces
Permissionless, agent-first access to isolated Linux workspaces.
Each tenant gets its own isolated, non-root workspace — a Cloudflare Sandbox microVM when the runtime is enabled — with an authenticated shell (PTY or exec), supported language runtimes, and HTTP services. Project files are meant to persist through documented snapshot/restore to object storage, not as a POSIX disk mount. What changes is the gate: an agent with a wallet should authenticate, pay a prepaid hourly term with x402 USDC on Base, and operate a workspace without a human opening a cloud account. A human dashboard uses the same API and is optional — not the required path.
This public prototype has a live SIWE login and request API. It does not yet take payment or start a workspace, it does not claim to be the first agentic cloud, and permissionless onboarding still needs usage rules and abuse controls. Non-root alone is not the isolation boundary.
Workspace catalog
Isolated workspaces, not full root SSH virtual machines. Prepaid hourly rental of a fixed size is the intended customer billing model. Published rates appear only after the Sandbox runtime and a verified x402 flow are connected; nothing here is a live price or a charge.
Loading catalog…
What is ready, and what is not
- Ready: an agent-readable API, OpenAPI schema, SIWE wallet authentication, workspace request records, budgets, and an audit log. Humans can use the optional dashboard, which calls the same endpoints.
- Not ready: x402 settlement, the Cloudflare Sandbox runtime, live PTY/exec, HTTP service exposure, or snapshot storage. Start, stop, destroy, exec, terminal, and any payment payload stay blocked and will not report a paid or provisioned success.
- Limits of the claim: this is an access upgrade, not a promise that upstream infrastructure is permissionless, or that there will be no usage rules. The control plane — including the Dubai development host — is not used to run tenant workloads. R2 is not a POSIX block device, and memory is not guaranteed across restart.
Optional human dashboard
Agents should start at the quickstart and
OpenAPI. This dashboard is a convenience
over the same API. Connect an injected wallet (EIP-6963 or
window.ethereum). Signing proves you own the account; it
does not approve a payment. WalletConnect is not configured, and a
listing inside any particular wallet app is not guaranteed.
Wallet
Not signed in.
Request a workspace
Your requests
Sign in to see requests.
Agent API
After SIWE, requests are scoped to the signing wallet. The start, exec,
and terminal endpoints are implemented and will refuse to provision until
the Sandbox runtime is enabled. x402 payment submissions are rejected
until settlement is enabled — they never return paid: true
today. Full narrative and examples:
/docs/agent.md,
/llms.txt,
/openapi.json.
GET /docs/agent.md
GET /llms.txt
GET /openapi.json
GET /api/runtime
GET /api/x402 payments not enabled; accepts: []
POST /api/x402/pay 409, paid=false, charged=false
POST /api/auth/nonce
POST /api/auth/verify { "message", "signature" }
GET /api/session
GET /api/catalog
GET /api/workspaces
POST /api/workspaces { "planId", "billingTerm", "name?", "budgetHours?" }
POST /api/workspaces/:id/start blocked, not a live sandbox
POST /api/workspaces/:id/exec 409, runtime not enabled
POST /api/workspaces/:id/terminal 409, not root SSH
POST /api/workspaces/:id/stop
POST /api/workspaces/:id/destroy
GET /api/workspaces/:id/events