Show HN: LunaRoute – a high-performance local proxy for AI coding assistants (github.com)

LunaRoute is a new high-performance local proxy for AI coding assistants (Claude Code, OpenAI Codex CLI, OpenCode) that gives developers full visibility into every LLM interaction while adding almost no latency. Run a single Rust binary and point your client at localhost:8081 to get zero-copy, sub-millisecond passthrough with “100% API fidelity” (preserves streaming/extended fields) and automatic auth handling from client headers or ~/.codex/auth.json. It supports Anthropic and OpenAI dialects simultaneously, can translate Anthropic requests to OpenAI-compatible endpoints (e.g., Gemini), and exposes request paths (/v1/messages and /v1/chat/completions) natively so tools receive responses in the expected format. Its value for ML engineers and teams is observability, cost control, and compliance: optional session recording (lightweight SQLite analytics or full JSONL logs), web UI for browsing sessions, token-level accounting, per-tool latency metrics, and search/replay of past conversations. Security features include automatic PII redaction with multiple modes (mask/remove/tokenize), HMAC-based deterministic tokenization for reversible redaction, and configurable retention. Built in Rust with ~0.1–0.5ms overhead in passthrough/async-recording modes, zero config defaults, and comprehensive tests (73% coverage, 544 passing), LunaRoute is positioned as a lightweight, production-ready proxy for teams that need debugging, unified logging, and regulatory-safe handling of AI interactions.