ShipClaw is managed hosting for Claude-powered AI agents built on OpenClaw, Anthropic's open-source agent runtime. Sign up is free with 25 trial credits; users pick from pre-built personalities or define their own with a single SOUL.md file, paste a Telegram bot token, and have a Claude-powered agent on Telegram in under 60 seconds. No SSH, no Docker, no Anthropic API key to manage. Pricing is pay-per-credit with no monthly subscription. The minimum paid top-up is a $15 credit pack (1500 credits at ~$0.01 per credit). Larger packs unlock bonus credits — 10% at $25, 20% at $50, 30% at $100, 40% at $200. Credits don't expire on a monthly cadence; you burn them down on real usage. Idle agents cost nothing. The platform is built on a shared-pool architecture: each container hosts up to ~50 Claude agents, with atomic Postgres-level user assignment to handle burst signups. Each user runs in an isolated sandbox with a 1.5 GB memory watchdog and a 15-minute idle timeout (1 hour for Telegram-bound agents). The bundled stealth browser engine lets agents browse the live web and bypass CAPTCHA-class checks. ShipClaw is Anthropic-only by design — single-model focus enables predictable unit economics for pay-per-credit billing. There is no bring-your-own-key option; ShipClaw holds the Anthropic API contract on the customer's behalf. OpenClaw itself is open source and self-hostable for users who prefer that path.