Introduction
White-label PaaS where tenants build businesses on top of AI agents.
HydraCore is a multi-tenant Platform-as-a-Service for AI agents. Tenants ship agents to their customers, packaged with a runtime, a brain, channels, and a credit-system primitive. The Entity Store is the tenant-defined data layer underneath every agent.
Three-layer composition
Every deployed instance is Runtime × Brain × Channels[]:
- Runtime — the long-running process on the tenant's VPS that takes a chat turn, calls the LLM, runs Skills, and reports usage. Today: OpenClaw (TS/Bun, default), Hermes (Python, experimental), goose (Rust, experimental).
- Brain — the agent's configuration: system prompt, model policy, skills, MCP servers, voice config, channel routing.
- Channels — the surfaces customers reach the agent on: WhatsApp, Voice (Pipecat / Retell / LiveKit), SMS, Telegram, Web chat, Email.
Two products
- Olympus — DIY: tenants compose recipes themselves and deploy via the wizard. CrewAmp, ChatWeb, and Leslie are Olympus-style.
- Elysium — turnkey verticals: HydraCore ships pre-composed recipes for specific industries. PropManager (AU/UK/US property management) is the current Elysium vertical.
Where to start
- Quickstart — deploy your first agent in under 10 minutes.
- Architecture — the composition model, primitives, and how they fit together.
- Runtimes — the runtime catalog (auto-generated from manifests on every build).