token-chai

Token-efficient LLM prompting via Chinese compression.

Translates your prompt to Chinese before sending to the AI (Chinese uses fewer tokens than English), then translates the response back to your language, whic will save you money on API costs.

Installation

pip install token-chai

Supported Models

Key	Provider	Default Model
openai	OpenAI	gpt-4o
google-genai	Google AI	gemini-2.5-flash
claude	Anthropic	claude-sonnet-4-20250514

---

Usage

There are two ways to use token-chai:

Method 1: Configure once, use everywhere (recommended)

Set your options once with chai.configure() and never repeat them:

import chai

chai.configure(
    ai_model="google-genai",
    api_key="your api key",
    model="gemini-2.5-flash",
    response_lang="en",
)

response = chai.get_response("Explain black holes in simple terms")
print(response)

You can still override any option per call:

response = chai.get_response("What is recursion?", response_lang="hi")

---

Method 2: Pass everything directly

import chai

response = chai.get_response(
    prompt="Explain black holes in simple terms",
    ai_model="google-genai",
    api_key="your api key",
    model="gemini-2.5-flash",
    response_lang="en",
)
print(response)

---

Examples

Google Gemini

Configure once:

import chai

chai.configure(
    ai_model="google-genai",
    api_key="your api key",
    model="gemini-2.5-flash",
    response_lang="en",
)

response = chai.get_response("What is the theory of relativity?")
print(response)

Direct:

import chai

response = chai.get_response(
    prompt="What is the theory of relativity?",
    ai_model="google-genai",
    api_key="your api key.",
    model="gemini-2.5-flash",
    response_lang="en",
)
print(response)

---

OpenAI ChatGPT

Configure once:

import chai

chai.configure(
    ai_model="openai",
    api_key="your api key",
    model="gpt-4o",
    response_lang="en",
)

response = chai.get_response("Explain quantum computing simply")
print(response)

Direct:

import chai

response = chai.get_response(
    prompt="Explain quantum computing simply",
    ai_model="openai",
    api_key="your api key",
    model="gpt-4o",
    response_lang="en",
)
print(response)

---

Anthropic Claude

Configure once:

import chai

chai.configure(
    ai_model="claude",
    api_key="your api key",
    model="claude-sonnet-4-20250514",
    response_lang="en",
)

response = chai.get_response("How does photosynthesis work?")
print(response)

Direct:

import chai

response = chai.get_response(
    prompt="How does photosynthesis work?",
    ai_model="claude",
    api_key="your api key",
    model="claude-sonnet-4-20250514",
    response_lang="en",
)
print(response)

---

Check Token Savings

Preview how much compression you get before sending:

import chai

stats = chai.estimate_savings("Explain black holes in simple terms")
print(stats)
# {
#   'original_chars': 36,
#   'chinese_chars': 17,
#   'compression_ratio': 0.47,
#   'chinese_preview': '用简单的术语解释黑洞...'
# }

---

Response Languages

Pass any BCP-47 language code to response_lang:

Code	Language
en	English
hi	Hindi
fr	French
de	German
ja	Japanese
es	Spanish
zh	Chinese (skip back-translation)

---

Debug Mode

Skip translation entirely to test raw API responses:

response = chai.get_response(
    prompt="Hello!",
    skip_translation=True,
)
print(response)

---

License

MIT