Stream validated output

To stream validated output, pass the stream=True flag through the guard function. This will return a generator that yields GuardResult objects as they are processed.

from guardrails import Guard
import os

# Set your OpenAI API key
# os.environ['OPENAI_API_KEY'] = ""

guard = Guard()

stream_chunk_generator = guard(
    messages=[{"role":"user", "content":"How many moons does Jupiter have?"}],
    model="gpt-3.5-turbo",
    stream=True
)

# Print the validated output as it is processed
for chunk in stream_chunk_generator:
    print(f"{chunk.validated_output}")

Using validators with streaming

Using validators with streaming works the same way. Note that not all on_fail types are supported with streaming. See the full list in the error remediation concepts doc.

pip install guardrails-ai-profanity-free

from guardrails import Guard
from guardrails_ai.profanity_free import ProfanityFree
import os

# Set your OpenAI API key
# os.environ['OPENAI_API_KEY'] = ""

guard = Guard().use(ProfanityFree())

stream_chunk_generator = guard(
    messages=[{"role":"user", "content":"How many moons does Jupiter have?"}],
    model="gpt-3.5-turbo",
    stream=True
)

# Print the validated output as it is processed
for chunk in stream_chunk_generator:
    print(f"{chunk.validated_output}")

Learn more

Read more about streaming in our concept docs:

How-to guides

Sample apps

Using validators with streaming

Learn more

​Using validators with streaming

​Learn more

Using validators with streaming

Learn more