​
Overview

The Upsonic framework can use .env variable files or environment variables for LLM support. Once you provide the keys for various services, you can easily select the LLM by using the model parameter within the agents. The supported LLMs are:
  • OpenAI
    • openai/gpt-4o
    • openai/gpt-4.5-preview
    • openai/o3-mini
    • openai/gpt-4o-mini
  • Azure
    • azure/gpt-4o
    • azure/gpt-4o-mini
  • Anthropic
    • claude/claude-3-5-sonnet
    • claude/claude-3-7-sonnet
  • AWS Bedrock
    • bedrock/claude-3-5-sonnet
  • DeepSeek
    • deepseek/deepseek-chat
  • Google Gemini
    • gemini/gemini-2.0-flash
    • gemini/gemini-1.5-pro
    • gemini/gemini-1.5-flash
  • Ollama
    • ollama/llama3.2
    • ollama/llama3.1:70b
    • ollama/llama3.1
    • ollama/llama3.3
    • ollama/qwen2.5
  • OpenRouter
    • openrouter/anthropic/claude-3.5-sonnet
    • openrouter/meta-llama/llama-3-70b
    • openrouter/google/gemini-pro
      • See the note below for more information about OpenRouter’s gateway functionality
Note: OpenRouter acts as a gateway to access various LLM models from different providers. You can use any model available on OpenRouter by following the pattern openrouter/{provider}/{model-name}. This allows you to access models that might not be directly supported through their native APIs.

​
Setting up .env

To use these LLMs, an example .env variable file is as follows. You can adjust the variables according to the LLM you want to use and fill in only the ones required. This .env file must be located in your working directory. 
# OpenAI
OPENAI_API_KEY="sk-***"

# Anthropic
ANTHROPIC_API_KEY="sk-***"

# DeepSeek
DEEPSEEK_API_KEY="sk-**"

# AWS Bedrock
AWS_ACCESS_KEY_ID="**"
AWS_SECRET_ACCESS_KEY="***"
AWS_REGION="**-**"

# Azure
AZURE_OPENAI_ENDPOINT="https://**.com/"
AZURE_OPENAI_API_VERSION="****-**-**"
AZURE_OPENAI_API_KEY="***"

# Google Gemini
GOOGLE_GLA_API_KEY="***"

# OpenRouter
OPENROUTER_API_KEY="***"

​
Using an specific model in Agent

Using the model parameter, you can easily select which LLM each agent will use. You can refer to the example below: 
from upsonic import Agent

product_manager_agent = Agent(
    name="Product Manager",
    model="openai/gpt-4o" # Specify the model
)

​
Using an specific model in Direct

As like Agent you can specify model directly with model parameter. 
from upsonic import Direct

direct = Direct(model="openai/gpt-4o")