langchain-maritaca 0.2.3

An integration package connecting Maritaca AI and LangChain for Brazilian Portuguese language models

langchain-maritaca

🇧🇷 Leia em Português

An integration package connecting Maritaca AI and LangChain for Brazilian Portuguese language models.

Author: Anderson Henrique da Silva Location: Minas Gerais, Brasil GitHub: anderson-ufrj

Overview

Maritaca AI provides state-of-the-art Brazilian Portuguese language models, including the Sabiá family of models. This integration allows you to use Maritaca's models seamlessly within the LangChain ecosystem.

Available Models

Model	Description	Pricing (per 1M tokens)
sabia-3.1.1	Most capable model, best for complex tasks	Check Maritaca AI for pricing
sabiazinho-3.1	Fast and economical, great for simple tasks	Check Maritaca AI for pricing

Installation

pip install langchain-maritaca

Setup

Set your Maritaca API key as an environment variable:

export MARITACA_API_KEY="your-api-key"

Or pass it directly to the model:

from langchain_maritaca import ChatMaritaca

model = ChatMaritaca(api_key="your-api-key")

Usage

Basic Usage

from langchain_maritaca import ChatMaritaca

model = ChatMaritaca(
    model="sabia-3.1",
    temperature=0.7,
)

messages = [
    ("system", "Você é um assistente prestativo especializado em cultura brasileira."),
    ("human", "Quais são as principais festas populares do Brasil?"),
]

response = model.invoke(messages)
print(response.content)

Streaming

from langchain_maritaca import ChatMaritaca

model = ChatMaritaca(model="sabia-3.1", streaming=True)

for chunk in model.stream("Conte uma história sobre o folclore brasileiro"):
    print(chunk.content, end="", flush=True)

Async Usage

import asyncio
from langchain_maritaca import ChatMaritaca

async def main():
    model = ChatMaritaca(model="sabia-3.1")
    response = await model.ainvoke("Qual é a receita de pão de queijo?")
    print(response.content)

asyncio.run(main())

With LangChain Expression Language (LCEL)

from langchain_maritaca import ChatMaritaca
from langchain_core.prompts import ChatPromptTemplate

model = ChatMaritaca(model="sabia-3.1")

prompt = ChatPromptTemplate.from_messages([
    ("system", "Você é um especialista em {topic}."),
    ("human", "{question}"),
])

chain = prompt | model

response = chain.invoke({
    "topic": "história do Brasil",
    "question": "Quem foi Tiradentes?"
})
print(response.content)

With Tool Calling (Function Calling)

from langchain_maritaca import ChatMaritaca
from langchain_core.tools import tool

@tool
def get_weather(city: str) -> str:
    """Get the current weather for a city."""
    return f"O clima em {city} está ensolarado, 25°C"

model = ChatMaritaca(model="sabia-3.1")
model_with_tools = model.bind_tools([get_weather])

response = model_with_tools.invoke("Como está o tempo em São Paulo?")
print(response)

Why Maritaca AI?

Maritaca AI models are specifically trained for Brazilian Portuguese, offering:

• Native Portuguese Understanding: Better comprehension of Brazilian idioms, expressions, and cultural context
• Local Data Training: Trained on diverse Brazilian Portuguese data sources
• Cost-Effective: Competitive pricing for Portuguese language tasks
• Low Latency: Servers located in Brazil for faster response times

Used in Production

Cidadão.AI - Brazilian government transparency platform powered by AI agents, handling 331K+ requests/month.

• Frontend: github.com/anderson-ufrj/cidadao.ai-frontend
• Backend: github.com/anderson-ufrj/cidadao.ai-backend

Using this package in production? Open an issue to get featured!

API Reference

ChatMaritaca

Main class for interacting with Maritaca AI models.

Parameters:

Parameter	Type	Default	Description
model	str	"sabia-3.1"	Model name to use
temperature	float	0.7	Sampling temperature (0.0-2.0)
max_tokens	int	None	Maximum tokens to generate
top_p	float	0.9	Top-p sampling parameter
api_key	str	None	Maritaca API key (or use env var)
base_url	str	"https://chat.maritaca.ai/api"	API base URL
timeout	float	60.0	Request timeout in seconds
max_retries	int	2	Maximum retry attempts
retry_if_rate_limited	bool	True	Auto-retry on rate limit (HTTP 429)
retry_delay	float	1.0	Initial delay between retries (seconds)
retry_max_delay	float	60.0	Maximum delay between retries (seconds)
retry_multiplier	float	2.0	Multiplier for exponential backoff
streaming	bool	False	Enable streaming responses

Development

Setup

# Clone the repository
git clone https://github.com/anderson-ufrj/langchain-maritaca.git
cd langchain-maritaca

# Install dependencies
pip install -e ".[dev]"

# Run tests
pytest

# Run linting
ruff check .
ruff format .

# Run type checking
mypy langchain_maritaca

Running Tests

# Unit tests only
pytest tests/unit_tests/

# Integration tests (requires MARITACA_API_KEY)
pytest tests/integration_tests/

# With coverage
pytest --cov=langchain_maritaca --cov-report=html

Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

1. Fork the repository
2. Create your feature branch (git checkout -b feature/amazing-feature)
3. Commit your changes (git commit -m 'feat: add amazing feature')
4. Push to the branch (git push origin feature/amazing-feature)
5. Open a Pull Request

Changelog

See CHANGELOG.md for a list of changes.

License

This project is licensed under the MIT License - see the LICENSE file for details.

Related Projects

• LangChain - Building applications with LLMs through composability
• Maritaca AI - Brazilian Portuguese language models