Use Supported LLMs

Guardrails has support for 100+ LLMs through its integration with LiteLLM. This integration is really useful because it allows the Guardrails call API to use the same clean interface that LiteLLM and OpenAI use. This means that you can use similar code to make LLM requests with Guardrails as you would with OpenAI.

To interact with a model, set the desired LLM API KEY such as the OPENAI_API_KEY and the desired model with the model property. Examples are below for some common ones.

OpenAI

Basic Usage

from guardrails import Guard

os.environ["OPENAI_API_KEY"] = "YOUR_OPEN_AI_API_KEY"

guard = Guard()

result = guard(
    messages=[{"role":"user", "content":"How many moons does Jupiter have?"}],
    model="gpt-4o",
)

print(f"{result.validated_output}")

Streaming

from guardrails import Guard

os.environ["OPENAI_API_KEY"] = "YOUR_OPEN_AI_API_KEY"

guard = Guard()

stream_chunk_generator = guard(
    messages=[{"role":"user", "content":"How many moons does Jupiter have?"}],
    model="gpt-4o",
    stream=True,
)

for chunk in stream_chunk_generator
    print(f"{chunk.validated_output}")

Tools/Function Calling

from pydantic import BaseModel, Field
from typing import List
from guardrails import Guard

os.environ["OPENAI_API_KEY"] = "YOUR_OPEN_AI_API_KEY"

class Fruit(BaseModel):
    name: str
    color: str

class Basket(BaseModel):
    fruits: List[Fruit]
    
guard = Guard.for_pydantic(Basket)

result = guard(
    messages=[{"role":"user", "content":"Generate a basket of 5 fruits"}],
    model="gpt-4o",
    tools=guard.json_function_calling_tool([]),
    tool_choice="required",
)

print(f"{result.validated_output}")

Anthropic

Basic Usage

from guardrails import Guard
import os

guard = Guard()

os.environ["ANTHROPIC_API_KEY"] = "your-api-key"

result = guard(
    messages=[{"role":"user", "content":"How many moons does Jupiter have?"}],
    model="claude-3-opus-20240229"
)

print(f"{result.validated_output}")

Streaming

from guardrails import Guard
import os

os.environ["ANTHROPIC_API_KEY"] = "your-api-key"

guard = Guard()

stream_chunk_generator = guard(
    messages=[{"role":"user", "content":"How many moons does Jupiter have?"}],
    model="claude-3-opus-20240229",
    stream=True,
)

for chunk in stream_chunk_generator
    print(f"{chunk.validated_output}")

Azure OpenAI

Basic Usage

from guardrails import Guard
import os
os.environ["AZURE_API_KEY"] = "" # "my-azure-api-key"
os.environ["AZURE_API_BASE"] = "" # "https://example-endpoint.openai.azure.com"
os.environ["AZURE_API_VERSION"] = "" # "2023-05-15"

guard = Guard()

result = guard(
    model="azure/<your_deployment_name>",
    messages=[{"role":"user", "content":"How many moons does Jupiter have?"}],
)

print(f"{result.validated_output}")

Streaming

from guardrails import Guard

os.environ["AZURE_API_KEY"] = "" # "my-azure-api-key"
os.environ["AZURE_API_BASE"] = "" # "https://example-endpoint.openai.azure.com"
os.environ["AZURE_API_VERSION"] = "" # "2023-05-15"

guard = Guard()

stream_chunk_generator = guard(
    messages=[{"role":"user", "content":"How many moons does Jupiter have?"}],
    model="azure/<your_deployment_name>", 
    stream=True
)

for chunk in stream_chunk_generator
    print(f"{chunk.validated_output}")

Tools/Function Calling

from pydantic import BaseModel, Field
from typing import List
from guardrails import Guard

os.environ["AZURE_API_KEY"] = "" # "my-azure-api-key"
os.environ["AZURE_API_BASE"] = "" # "https://example-endpoint.openai.azure.com"
os.environ["AZURE_API_VERSION"] = "" # "2023-05-15"

class Fruit(BaseModel):
    name: str
    color: str

class Basket(BaseModel):
    fruits: List[Fruit]
    
guard = Guard.for_pydantic(Basket)

result = guard(
    messages=[{"role":"user", "content":"Generate a basket of 5 fruits"}],
    model="azure/<your_deployment_name>", 
    tools=guard.add_json_function_calling_tool([]),
    tool_choice="required",
)

print(f"{result.validated_output}")

Gemini

Basic Usage

from guardrails import Guard
import os

os.environ['GEMINI_API_KEY'] = ""
guard = Guard()

result = guard(
    messages=[{"role":"user", "content":"How many moons does Jupiter have?"}],
    model="gemini/gemini-pro"
)

print(f"{result.validated_output}")

Streaming

from guardrails import Guard
import os

os.environ['GEMINI_API_KEY'] = ""
guard = Guard()
stream_chunk_generator = guard(
    messages=[{"role":"user", "content":"How many moons does Jupiter have?"}],
    model="gemini/gemini-pro",
    stream=True
)

for chunk in stream_chunk_generator
    print(f"{chunk.validated_output}")

COMING SOON - Tools/Function calling

from pydantic import BaseModel, Field
from typing import List
from guardrails import Guard

os.environ['GEMINI_API_KEY'] = ""

class Fruit(BaseModel):
    name: str
    color: str

class Basket(BaseModel):
    fruits: List[Fruit]
    
guard = Guard.for_pydantic(Basket)

result = guard(
    messages=[{"role":"user", "content":"Generate a basket of 5 fruits"}],
    model="gemini/gemini-pro"
    tools=guard.add_json_function_calling_tool([])
)

print(f"{result.validated_output}")

Databricks

Basic Usage

from guardrails import Guard

os.environ["DATABRICKS_API_KEY"] = "" # your databricks key
os.environ["DATABRICKS_API_BASE"] = "" # e.g.: https://abc-123ab12a-1234.cloud.databricks.com

guard = Guard()

result = guard(
    messages=[{"role":"user", "content":"How many moons does Jupiter have?"}],
    model="databricks/databricks-dbrx-instruct",
)

print(f"{result.validated_output}")

Streaming

from guardrails import Guard

os.environ["DATABRICKS_API_KEY"] = "" # your databricks key
os.environ["DATABRICKS_API_BASE"] = "" # e.g.: https://abc-123ab12a-1234.cloud.databricks.com

guard = Guard()

stream_chunk_generator = guard(
    messages=[{"role":"user", "content":"How many moons does Jupiter have?"}],
    model="databricks/databricks-dbrx-instruct",
    stream=True,
)

for chunk in stream_chunk_generator
    print(f"{chunk.validated_output}")

Other LLMs

See LiteLLM’s documentation here for details on many other llms.

Custom LLM Wrappers

In case you're using an LLM that isn't natively supported by Guardrails and you don't want to use LiteLLM, you can build a custom LLM API wrapper. In order to use a custom LLM, create a function that accepts a positional argument for the prompt as a string and any other arguments that you want to pass to the LLM API as keyword args. The function should return the output of the LLM API as a string. Install ProfanityFree from hub:

guardrails hub install hub://guardrails/profanity_free

from guardrails import Guard
from guardrails.hub import ProfanityFree

# Create a Guard class
guard = Guard().use(ProfanityFree())

# Function that takes the prompt as a string and returns the LLM output as string
def my_llm_api(
    *,
    **kwargs
) -> str:
    """Custom LLM API wrapper.

    At least one of messages should be provided.

    Args:
        **kwargs: Any additional arguments to be passed to the LLM API

    Returns:
        str: The output of the LLM API
    """
    messages = kwargs.pop("messages", [])
    updated_messages = some_message_processing(messages)
    # Call your LLM API here
    # What you pass to the llm will depend on what arguments it accepts.
    llm_output = some_llm(updated_messages, **kwargs)

    return llm_output

# Wrap your LLM API call
validated_response = guard(
    my_llm_api,
    messages=[{"role":"user","content":"Can you generate a list of 10 things that are not food?"}],
    **kwargs,
)