Spooky Labs

Spooky Labs is a platform for creating and operating headless AI agents. Configure a prompt, model, skills, delegation, paper-trading markets and the schedules that run the agent; the platform starts every run from a schedule, streams its progress and records usage for billing.

The platform

  • Runtime: pinned kagent alpha5 Agent, AgentTemplate and Harness resources, reconciled by the kagent chart’s own controller, with sessions on tenant Substrate WorkerPools. Skills are baked into the custom runtime; native Bash is removed.
  • API: a Go Connect server behind the GKE Gateway, enforcing identity, tenant boundaries and platform workflows for the two browser consoles.
  • Models: an in-cluster agentgateway path to Vertex AI, with attributed usage telemetry.
  • Billing: gateway access telemetry (tokens) and Substrate actor lifecycle records (running seconds) feed a durable ledger and Metronome usage rating. Usage is invoiced monthly and charged to the card on file, which is also charged each time a month’s uncharged usage reaches $100. There is no prepaid balance and no hard stop. Compute is billed from the pinned bill-from, 2026-09-30 05:00 UTC. See the measurement limits in Metering and billing.
  • Brokerage: Alpaca sandbox accounts and market-scoped tools. Balances, positions and orders are paper money.

The hosts

Host Purpose
spookylabs.ai Marketing placeholder from www; unknown paths redirect to the developer host
developer.spookylabs.ai The agent console from web (the developer preview)
admin.spookylabs.ai Operator console from admin
docs.spookylabs.ai This documentation site
api.spookylabs.ai Connect, gRPC and gRPC-Web from platform-api

Reading these pages

These pages describe the platform at platform-api df29c41 and iac 2ce48f4, except billing, which describes the monthly invoicing that replaced prepaid credits (api #85), and headless agents, which describe the contract in api #87: no chat, and every run starts from a schedule. A source implementation and a successful local test are not evidence of a deployed rollout: check the deployed image before relying on a newly changed behavior.

The pinned runtime is the API authority. Current upstream documentation has already advanced to AgentInstances, while this integration still uses the alpha5 Agent resource. Unimplemented, configuration-gated and incomplete features are separated on Availability and limits.