Iudex

Next generation observability.

Supported Libraries

✅ logger ✅ loguru ✅ sqlalchemy ✅ supabase ✅ modal ✅ streamlit

Supported libraries from OTel:

✅ aio-pika ✅ aiohttp-client ✅ aiohttp-server ✅ aiopg ✅ asgi ✅ asyncio ✅ asyncpg ✅ aws-lambda ✅ boto ✅ boto3sqs ✅ botocore ✅ cassandra ✅ celery ✅ confluent-kafka ✅ dbapi ✅ django ✅ elasticsearch ✅ falcon ✅ fastapi ✅ flask ✅ grpc ✅ httpx ✅ jinja2 ✅ kafka-python ✅ logging ✅ mysql ✅ mysqlclient ✅ pika ✅ psycopg ✅ psycopg2 ✅ pymemcache ✅ pymongo ✅ pymysql ✅ pyramid ✅ redis ✅ remoulade ✅ requests ✅ sklearn ✅ sqlalchemy ✅ sqlite3 ✅ starlette ✅ system-metrics ✅ threading ✅ tornado ✅ tortoiseorm ✅ urllib ✅ urllib3 ✅ wsgi

Supported libraries from OpenLLMetry:

✅ OpenAI / Azure OpenAI ✅ Anthropic ✅ Cohere ✅ Ollama ✅ Mistral AI ✅ HuggingFace ✅ Bedrock (AWS) ✅ Replicate ✅ Vertex AI (GCP) ✅ Google Generative AI (Gemini) ✅ IBM Watsonx AI ✅ Together AI ✅ Aleph Alpha ✅ Chroma ✅ Pinecone ✅ Qdrant ✅ Weaviate ✅ Milvus ✅ Marqo ✅ LangChain ✅ LlamaIndex ✅ Haystack ✅ LiteLLM

Table of contents

• Iudex
  • Supported Libraries
  • Table of contents
• Getting Started
  • Autoinstrumentation (Most Common)
  • Log Attributes
  • Trace Span Attributes
• Integrations
  • Django
  • Modal
  • Streamlit
    • Using a main function
  • Tracing Your Functions
• Slack Alerts

Getting Started

Instrumenting your Python code with Iudex just takes a few steps.

1. Install dependencies.

pip install iudex

2. Follow the below instructions for your frameworks or use autoinstrumentation.
3. Make sure your app has access to the environment variable IUDEX_API_KEY. You can manually add this to instrument as well if you use something like a secrets manager.
4. You should be all set! Go to https://app.iudex.ai/ and enter your API key.
5. Go to https://app.iudex.ai/logs and press Search to view your logs.

Autoinstrumentation (Most Common)

Add this code to the VERY TOP of your entrypoint file, before all imports.

from iudex import instrument
instrument(
    service_name="YOUR_SERVICE_NAME", # highly encouraged
    env="prod", # dev, local, etc
    iudex_api_key="WRITE_ONLY_IUDEX_KEY", # only ever commit your WRITE ONLY key
)
# ^ Must run above all imports

Iudex auto-instrumentation must run before imports in order to patch libraries with specialized, no-overhead instrumentation code.

You should be all set! Iudex will now record logs and trace the entire life cycle for each request.

Go to https://app.iudex.ai/ to start viewing your logs and traces!

Log Attributes

You can add custom attributes to your logs by passing a dictionary to the extra parameter of the logging functions.

logger.info("Hello Iudex!", extra={"my_custom_attribute": "my_custom_value"})

These attributes will be searchable and displayed on your logs in the Iudex dashboard.

Trace Span Attributes

You can add custom attributes to the current trace span (if one exists) as follows:

from iudex import set_attribute
# ... inside some function/span
set_attribute(key="my_custom_attribute", value="my_custom_value")
# ... rest of function

These attributes will be searchable and displayed on your trace spans in the Iudex dashboard.

Integrations

Some frameworks are auto-instrumented through different entrypoints. Find your framework below and follow the instructions to get set up. (If it's not listed, the above section will work just fine!)

Django

Add the following code to your manage.py file in the main function entrypoint.

from iudex import instrument

def main():
    """Run administrative tasks."""
    os.environ.setdefault('DJANGO_SETTINGS_MODULE', 'mysite.settings')

    instrument()

    # ...

Modal

Because Modal only loads libraries after running a particular serverless function, you need follow their suggestions for using packages.

For example:

@app.function(image=image, secrets=[modal.Secret.from_name("iudex-api-key")])
def square(x):
    import logging
    from iudex import instrument

    instrument(service_name="test-modal-instrumentation", env="development")

    logging.basicConfig(level=logging.INFO)
    logger = logging.getLogger(__name__)

    logger.info("This log shows up on Iudex!")
    return x**2

Streamlit

Streamlit works by running a static Python script which means all you need to do is encapsulate the scripts contents.

1. Import iudex and call instrument() to the top of your Streamlit app file.
2. Add start_trace and end_trace to the top and bottom of the file, respectively.

from iudex import instrument, start_trace, end_trace, trace
instrument(
    service_name="YOUR_SERVICE_NAME", # highly encouraged
    env="prod", # dev, local, etc
    iudex_api_key="WRITE_ONLY_IUDEX_KEY", # only ever commit your WRITE ONLY key
)
# ^ Must run above all imports
import streamlit as st

# call start_trace before your Streamlit app logic
token = start_trace(name="streamlit-app")

# your Streamlit app logic...
def generate_text(topic: str, mood: str = "", style: str = ""):
    pass

end_trace(token)
# bottom of the file

3. Add the trace decorator to functions that you want to track. This will show the function and its arguments in the stacktrace.

@trace
def generate_text(topic: str, mood: str = "", style: str = ""):
    ...

Using a main function

If your entire Streamlit app runs a single function, then the setup is a bit easier.

1. Import iudex and call instrument() to the top of your Streamlit app file.
2. Add the trace decorator to the main function.

from iudex import instrument, trace
instrument(
    service_name="YOUR_SERVICE_NAME", # highly encouraged
    env="prod", # dev, local, etc
    iudex_api_key="WRITE_ONLY_IUDEX_KEY", # only ever commit your WRITE ONLY key
)

# ^ Must run above all imports
import streamlit as st

@trace
def main():
    # your Streamlit app logic...

main()

3. We still recommend adding the trace decorator to various functions.

Tracing Your Functions

Iudex automatically traces framework and library functions. For instance, if you use a framework like FastAPI or Django, Iudex will record subsequent library calls, durations, and metadata throughout the lifecycle of each request.

That said, you can also get more granular and trace your own functions. A good heuristic for "when should I trace my own function" is whenever it might be helpful to see the function's stack trace.

Note: You must call iudex.instrument() earlier in your code (per above) before the traced function is invoked.

from iudex import instrument, trace
instrument(
    service_name="YOUR_SERVICE_NAME", # highly encouraged
    env="prod", # dev, local, etc
    iudex_api_key="WRITE_ONLY_IUDEX_KEY", # only ever commit your WRITE ONLY key
)
# ^ Must run above all imports

@trace
def my_function(arg1, arg2):
  pass

Slack Alerts

You can easily configure Slack alerts on a per-log basis.

First visit https://app.iudex.ai/logs and click on the Add to Slack button in the top right.

Once installed to your workspace, tag your logs with the iudex.slack_channel_id attribute.

logger.info("Hello from Slack!", extra={"iudex.slack_channel_id": "YOUR_SLACK_CHANNEL_ID"})

Your channel ID can be found by clicking the name of the channel in the top left, then at the bottom of the dialog that pops up.

As long as the channel is public or you've invited the Iudex app, logs will be sent as messages to their tagged channel any time they are logged.