llmproxy

Protocol-translating LLM proxy for coding agents.

The first supported mode is claude2codex: clients speak OpenAI Responses API and upstream providers speak Anthropic Messages API. This lets Codex and OpenAI Agents use Claude-compatible providers.

Run

go run ./cmd/llmproxy

The service listens on :8080 by default. Override with PORT or LLMPROXY_ADDR.

URL Shape

Take the provider's Anthropic-compatible base URL:

https://api.z.ai/api/anthropic

Put https://llmproxy.sandbox0.ai/claude2codex/ in front of it:

https://llmproxy.sandbox0.ai/claude2codex/https://api.z.ai/api/anthropic

The proxy service handles protocol routes only. Hosted marketing pages and public ingress can be managed by a separate website deployment that forwards proxy paths to this service.

Environment

Variable	Description
PORT	HTTP port when LLMPROXY_ADDR is unset.
LLMPROXY_ADDR	Full listen address, for example :8080.

Current Status

Implemented:

• claude2codex for Anthropic-compatible base URLs
• OpenAI Responses text input to Anthropic Messages
• Function tool call and tool result conversion
• OpenAI web_search tool to Anthropic web_search_20250305 server tool conversion
• Anthropic Messages text/tool output to OpenAI Responses
• SSE response shape for streaming clients
• SSRF hardening for dynamic upstream URLs

Planned:

• codex2claude
• /responses/compact
• richer multimodal mapping
• persisted usage logging and per-key policy