legacyllm

Stop updating model names manually. legacyllm automatically detects deprecated AI models and routes them to their current replacements — across OpenAI, Anthropic, and Google Gemini.

import legacyllm

# claude-3-opus is retired — legacyllm swaps it automatically
response = legacyllm.chat(
    model="claude-3-opus-20240229",
    messages=[{"role": "user", "content": "Say hi"}]
)

# [legacyllm] 'claude-3-opus-20240229' is deprecated, switching to 'claude-opus-4-8'
print(response["text"])        # Hi!
print(response["model_used"])  # claude-opus-4-8
print(response["was_swapped"]) # True
print(response["usage"])       # {"input_tokens": 10, "output_tokens": 4, "total_tokens": 14}

---

Install

pip install legacyllm

Set your API keys in a .env file:

OPENAI_API_KEY=...
ANTHROPIC_API_KEY=...
GEMINI_API_KEY=...

---

Usage

Basic

import legacyllm

response = legacyllm.chat(
    model="claude-opus-4-6",
    messages=[{"role": "user", "content": "Hello"}]
)

print(response["text"])

With parameters

response = legacyllm.chat(
    model="gpt-4o",
    messages=[{"role": "user", "content": "Hello"}],
    max_tokens=512,
    temperature=0.7,
    system="You are a helpful assistant."
)

Conversation history

response = legacyllm.chat(
    model="claude-opus-4-6",
    messages=[
        {"role": "user", "content": "My name is John"},
        {"role": "assistant", "content": "Hi John!"},
        {"role": "user", "content": "What's my name?"}
    ]
)

Streaming

for chunk in legacyllm.chat(
    model="gpt-4o",
    messages=[{"role": "user", "content": "Tell me a story"}],
    stream=True
):
    print(chunk, end="", flush=True)

Async

import asyncio
import legacyllm

async def main():
    response = await legacyllm.async_chat(
        model="gemini-2.5-flash",
        messages=[{"role": "user", "content": "Hello"}]
    )
    print(response["text"])

asyncio.run(main())

Tools / function calling

response = legacyllm.chat(
    model="gpt-4o",
    messages=[{"role": "user", "content": "What's the weather in NYC?"}],
    tools=[{
        "type": "function",
        "name": "get_weather",
        "description": "Get weather for a location",
        "parameters": {
            "type": "object",
            "properties": {
                "location": {"type": "string"}
            },
            "required": ["location"]
        }
    }]
)

---

Response

Every response returns the same shape regardless of provider:

{
    "text":           str,   # the response text
    "model_used":     str,   # actual model used (may differ if swapped)
    "was_swapped":    bool,  # True if model was deprecated and swapped
    "original_model": str,   # the model name you passed in
    "usage": {
        "input_tokens":  int,
        "output_tokens": int,
        "total_tokens":  int
    }
}

---

Supported providers

Provider	Models
OpenAI	gpt-*, o1, o3, o4
Anthropic	claude-*
Google	gemini-*

---

Helper commands

# see available parameters for a provider
legacyllm.params("openai")
legacyllm.params("anthropic")
legacyllm.params("google")

# list all current models
legacyllm.models()
legacyllm.models("openai")

# from the terminal
python -m legacyllm params openai
python -m legacyllm models
python -m legacyllm update

---

Keeping models up to date

legacyllm ships with a built-in deprecation index. To refresh it:

python -m legacyllm update

This fetches the latest models from all 3 providers, updates legacyllm/data/current_models.json, and flags any models that have disappeared so you can update the deprecation index.

---

Contributing

The deprecation index (legacyllm/data/depricated.json) needs ongoing maintenance as providers release and retire models. If you notice a missing deprecation:

1. Fork the repo
2. Add the entry to legacyllm/data/depricated.json with replacement and deprecated_on
3. Run pytest to validate
4. Submit a pull request

---

License

MIT