Deploy an Agent Pod
Deploys a hosted agent. The paid deploy rides the audited /compute/provision money path, so payment works exactly like provisioning a machine: pay from prepaid credits (use_credits: true) or complete an x402/MPP payment (a 402 challenge is returned when no payment credential is supplied). All /compute/provision fields (plan, region, prepaid_hours, network, ssh_public_key, …) are accepted alongside the pod-specific fields below; os_id is ignored (pods always run Ubuntu 24.04). Authenticate with the same compute auth as /compute/* (API key, session bearer, or wallet signature).
Headers
Compute API key. Alternatively use a session bearer or wallet-signature headers.
"x402c_..."
Body
Agent to deploy (see GET /pods/catalog).
"openclaw"
managed = we run the LLM (metered, tier-driven). byok = the agent uses your own OpenAI-compatible endpoint.
managed, byok Compute plan id from GET /compute/plans. DigitalOcean plans use the do: prefix.
Region id from GET /compute/regions.
Required when ai_mode is managed. A tier id from GET /pods/catalog (drives the model + machine floor).
"starter"
Optional model slug. Managed: honored only if it is in the chosen tier's list, else the tier default. BYOK: any slug.
Optional. Validated against the agent's (and tier's) RAM floor.
Pay from prepaid USD credits instead of an x402 wallet payment.
x402 payment network. MPP clients use Authorization: Payment and omit this.
base, solana, megaeth, robinhood Prepaid runtime in hours (minimum 24).
x >= 24Required when ai_mode is byok. Your provider API key (stored encrypted).
BYOK only. Your OpenAI-compatible base URL. Defaults to https://openrouter.ai/api/v1.
BYOK only. Provider API type (e.g. openai-completions).
Optional channel tokens, keyed by channel id (e.g. { "telegram": "<bot-token>" }). Must be supported by the agent.
BYOK memory options.
Optional. Rarely needed — pods are headless.
