hero_gpu

OpenRPC gateway in front of an SGLang Python backend, exposing OpenAI- and Anthropic-compatible endpoints, an installer, an SDK, and an admin UI.

Crates

• hero_gpu — CLI binary (handles --start / --stop, registers with hero_proc)
• hero_gpu_server — JSON-RPC server + OpenRPC spec + SGLang client
• hero_gpu_api — OpenAI + Anthropic HTTP adapters (library)
• hero_gpu_sdk — generated client from openrpc.json
• hero_gpu_installer — uv-based sandboxed Python/SGLang installer
• hero_gpu_ui — Askama + Bootstrap admin dashboard
• hero_gpu_examples — examples + integration test skeleton

Sockets

All under ~/hero/var/sockets/hero_gpu/:

• rpc.sock — JSON-RPC 2.0 (consumed by hero_gpu_sdk)
• openapi.sock — OpenAI + Anthropic compatible HTTP (consumed by external clients)
• ui.sock — admin dashboard

Quickstart

make install          # build + install binaries to ~/hero/bin/
make download-assets  # fetch Bootstrap, Icons, Unpoly into hero_gpu_ui/static/
make run              # start service via hero_proc
make logs             # tail server logs
make stop             # stop service

SGLang backend

Launched as a separate hero_proc action via scripts/sglang_runner.sh, running inside a uv-managed Python venv at ~/hero/var/hero_gpu/python/. The Rust server talks to it over 127.0.0.1:30000.

See crates/hero_gpu_server/openrpc.json for the full RPC interface.