proxyllm-client

Client helper for ProxyLLM, the OpenAI-compatible LLM gateway. One routing key rides provider lanes with fallback (a flat-fee ChatGPT/Codex subscription, self-hosted Claude Code bridges, metered API keys), budgets, and request logs.

pip install proxyllm-client[openai]

Point the OpenAI SDK at the gateway:

import proxyllm_client

client = proxyllm_client.OpenAI(api_key="pllm_...")
client.chat.completions.create(model="flagship", messages=[{"role": "user", "content": "hi"}])

Or set two environment variables and change nothing else: OPENAI_BASE_URL=https://api.proxyllm.ai/v1, OPENAI_API_KEY=pllm_....

Agents can create the account themselves (email OTP, no captcha or browser):

proxyllm_client.signup("operator@example.com")          # code lands in the inbox
r = proxyllm_client.verify("operator@example.com", "123456")
token = r["account_token"]                               # sk_..., returned once

state = proxyllm_client.account(token)                   # 402 payload until a human pays:
print(state["payment"]["checkout_url"])                 # relay this to your operator
# poll account(token) until state["plan"] == "pro", then:
key = proxyllm_client.create_routing_key(token, label="agent")["raw_key"]

Unpaid accounts answer HTTP 402 with the checkout link, the flat price ($129/mo), and the 48-hour removal deadline; error bodies are returned as values because they contain the next step. Free helpers: models(), key_info(routing_key), savings(monthly_bill).

Full agent flow: https://proxyllm.ai/auth.md. OpenAPI: https://proxyllm.ai/openapi.json. MCP server: npx -y proxyllm-mcp.