gate

Hellas Gate is a user-aligned llm inference server/proxy/router

in-progress

some of most of this exists, read as a wishlist rather than supported though

• Private local inference via built-in catgrad engine
• End-to-end encrypted peer-to-peer mesh overlay to aggregate capacity across networked nodes
• Use LetsEncrypt to generate a private cert for a public https://<your-node-id>.private.hellas.ai endpoint, Hellas will relay it over p2p
• Upstream support for any openai/anthropic-compatible providers, openrouter, vllm, ollama etc
• Supports Cursor, Codex and Claude-Code, open-webui, charm, etc via local http server
• API key management, cost-tracking, rate-limiting, quotas, permissions, etc
• Smart router can optimize routing for cost, latency, etc via 'virtual models'
• Capture/Log/Export all requests, responses, metadata through the gateway

future

• Once catgrad ZK backend is implemented, we can support verifying responses- check request was serviced correctly without quantization, context injection, tampered weights, etc
• Once protoproto consensus protocol is implemented, we can support settlement and thus create decentralized, trustless and permissionless markets for llm inference