OpenAI SDK Compatibility Guide

TokenPak exposes an OpenAI-compatible endpoint at /v1/chat/completions. If you're migrating from the OpenAI API or using tools that target the OpenAI format (LangChain ChatOpenAI, LiteLLM, Vercel AI SDK, etc.), point them at the TokenPak proxy with no code changes.

---

How It Works

Your app (openai SDK) → POST /v1/chat/completions → TokenPak proxy → Anthropic API

TokenPak translates the OpenAI Chat Completions format to Anthropic's Messages API, applies compression + caching, then translates the response back.

---

Python — openai SDK

from openai import OpenAI

client = OpenAI(
    base_url="http://localhost:8766/v1",
    api_key="your-anthropic-api-key",  # forwarded to Anthropic
)

response = client.chat.completions.create(
    model="claude-sonnet-4-6",          # Anthropic model via TokenPak routing
    messages=[
        {"role": "user", "content": "What is TokenPak?"}
    ],
)
print(response.choices[0].message.content)

Streaming

with client.chat.completions.create(
    model="claude-sonnet-4-6",
    messages=[{"role": "user", "content": "Explain caching in one paragraph."}],
    stream=True,
) as stream:
    for chunk in stream:
        delta = chunk.choices[0].delta.content or ""
        print(delta, end="", flush=True)

---

curl

curl http://localhost:8766/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer your-anthropic-api-key" \
  -d '{
    "model": "claude-sonnet-4-6",
    "messages": [{"role": "user", "content": "Hello from curl!"}]
  }'

Streaming via curl

curl http://localhost:8766/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer your-anthropic-api-key" \
  -d '{
    "model": "claude-sonnet-4-6",
    "messages": [{"role": "user", "content": "Count to 5."}],
    "stream": true
  }' --no-buffer

---

LangChain ChatOpenAI

from langchain_openai import ChatOpenAI

llm = ChatOpenAI(
    base_url="http://localhost:8766/v1",
    api_key="your-anthropic-api-key",
    model="claude-sonnet-4-6",
)
result = llm.invoke("What is 2 + 2?")
print(result.content)

---

Model Names

Pass Anthropic model IDs directly — TokenPak routes them to Anthropic:

Model string	Routes to
claude-sonnet-4-6	Anthropic claude-sonnet-4-6
claude-haiku-4-5	Anthropic claude-haiku-4-5
claude-opus-4-6	Anthropic claude-opus-4-6

TokenPak aliases (e.g. tokenpak-anthropic/claude-sonnet-4-6) are also accepted.

---

Supported Parameters

Parameter	Supported	Notes
model	✅	Anthropic model IDs
messages	✅	Full role/content array
stream	✅	SSE streaming
max_tokens	✅	Passed through
temperature	✅	Passed through
system (top-level)	✅	Converted to Anthropic system prompt
tools / tool_choice	✅	Passed through
n > 1	❌	Only n=1 supported
logprobs	❌	Not supported by Anthropic

---

Notes

• The proxy runs at http://localhost:8766 by default. Change via TOKENPAK_PORT.
• Your API key is forwarded to Anthropic — use your real ANTHROPIC_API_KEY.
• Response format matches OpenAI Chat Completions shape (.choices[0].message.content).
• See Getting Started for proxy setup.