Skip to content

cli-proxy.md

Latest commit

 

History

History
61 lines (47 loc) · 1.59 KB

cli-proxy.md

File metadata and controls

61 lines (47 loc) · 1.59 KB
Raw

CLI proxy mode

keymesh can run as a local HTTP proxy when you don't want to embed the library in code.

Start

# OpenAI (default adapter), reads keys from OPENAI_API_KEYS
export OPENAI_API_KEYS="sk-key1,sk-key2,sk-key3"
npx @takk/keymesh start --port 8787

# Anthropic
export ANTHROPIC_API_KEYS="sk-ant-key1,sk-ant-key2"
npx @takk/keymesh start --port 8788 --adapter anthropic

# Gemini
export GEMINI_API_KEYS="key1,key2"
npx @takk/keymesh start --port 8789 --adapter gemini

# Any REST API
export API_KEYS="tvly-key1,tvly-key2"
npx @takk/keymesh start --port 8790 --adapter http \
  --base-url https://api.tavily.com \
  --keys-env API_KEYS

Use

# From your app, point your OpenAI SDK at localhost
curl http://localhost:8787/v1/chat/completions \
  -H "content-type: application/json" \
  -d '{
    "model": "gpt-4.1",
    "messages": [{"role":"user","content":"Hello"}]
  }'

keymesh injects the rotated key as the Authorization (or x-api-key, or x-goog-api-key, depending on the adapter) header before forwarding upstream.

Inspect

# Live pool state via HTTP
curl http://localhost:8787/__keymesh_inspect | jq

# Or from the state file
npx @takk/keymesh inspect --state-file ./.keymesh-state.jsonl

Notes

  • The proxy is a thin pass-through for JSON request/response bodies. For streaming SSE responses (e.g., stream: true with OpenAI), use the library API in 1.0; streaming proxy lands in 1.1.
  • Multiple keymesh start instances can run on different ports for different providers; they don't share state unless you point them at the same --state-file.