i.AI Utility Code

i-dot-ai-utilities is a python package used and developed by the i.AI team within DSIT. It provides common features used in many of our applications.

Installation

When installing the package, the base package comes with only the logger module, to install more use extras. The following extras are available:

• auth
• file_store
• litellm
• metrics
• otel (pre-release)
• otel_django (pre-release)
• otel_fastapi (pre-release)
• django (pre-release)
• all

To install the package, use your package manager of choice:

pip install "i-dot-ai-utilities[all]"

poetry add "i-dot-ai-utilities[all]"

uv pip install "i-dot-ai-utilities[all]"

Replace [all] with any extras from the list above, comma separated, or remove entirely to install just the base package.

⚠️ The otel, otel_django, otel_fastapi, and django extras are pre-release, use with caution. They're compatible only with the OpenTelemetry PoC pipeline, not the existing i.AI observability stack. Pass --pre (or pin the pre-release version) to install them; a plain pip install won't pull a pre-release. [all] pulls them in too, so it also needs --pre during the pre-release window.

Features

Current features:

Structured Logging

The structured logging library is used to generate logs in a known format so they can be further processed into logging systems downstream. It also provides the ability to easily enrich log messages with useful data, and in some cases does this automatically.

You can find information on usage of the logging library in the logging library readme.

Metrics Collection

The metrics collection library provides the ability to write time-series metrics out to useful destinations. In the case of i.AI, this is CloudWatch Metrics.

There's also a handy interface provided which can be used in your code to allow for modularity if the swapping out of implementations is desired.

You can find information on usage of the metrics collection library in the metrics library readme.

File store

The file store library currently only supports aws s3 as this is the main/only file store that we use in anger.

It can be used to upload and download files, and generate file download links for end-users to use.

The aim is to be able to plug more file storage destinations into this module so it can be swapped out easily.

You can find out information on usage of the file store library in the file store library readme.

LiteLLM

This library currently supports LLM proxy through LiteLLM, for chat and embedding functions.

The hope for this library is to easily swap between proxies for whichever is best-in-market at the time.

As the end-user, you'll have to make sure that the API key issued to you by LiteLLM will support the models you're trying to use.

More information on usage and setup can be found in the litellm library readme.

OpenTelemetry & Django middleware

The otel extra adds a framework-agnostic OpenTelemetry bootstrap (configure_otel and the framework helpers configure_otel_for_django / configure_otel_for_fastapi / configure_otel_for_lambda) that wires traces, metrics, and an optional structlog log bridge to an OTLP endpoint, with W3C + AWS X-Ray propagation. Add otel_django or otel_fastapi for the matching auto-instrumentation, so you only pull the instrumentation you actually use.

The django extra adds request-lifecycle middleware: StructuredLoggingMiddlewareOTel for per-request structured logging and DjangoUserIdMiddleware for authenticated-user attribution. For a Django app wanting the full stack, install [django,otel_django].

⚠️ Pre-release, use with caution (see the Installation note): compatible only with the OpenTelemetry PoC pipeline, not the existing i.AI observability stack.

You can find usage details in the logging library readme.

Future features:

• authorisation
• vector stores

How to use

Unit Testing

All modules contained within this repo include robust test suites. You can run tests for all modules in this package using make test.

Tests and linting runs on every push and merge to main.

When making changes or adding tests, please ensure tests run in isolation, as failures of external dependencies will impact the CI tests for all packages. Please also make sure that tests pass before merging, as failing tests will impact every package in the application.

CI/CD & Releases

Package versions are published to PyPI by the Publish python package workflow. Version names must adhere to semantic versioning and must not be prefixed with a v. Merging to main does not publish anything; publishing is always a deliberate action.

There are three publish channels:

Notes:

• Manual dispatch is pre-release only. The workflow rejects a manual production publish whose version is not a PEP 440 pre-release (an rc / a / b / .dev suffix). Cut stable releases through the GitHub Release flow.
• The tag must already exist before a manual dispatch. The workflow checks out the tag you pass; create and push it first.

Installing a pre-release

From production PyPI (pre-release → production channel): pip will not resolve a PEP 440 pre-release unless you opt in with --pre or pin the exact version, so ordinary consumers stay on the latest stable automatically.

# opt in to pre-releases (resolves the newest rc)
pip install --pre "i-dot-ai-utilities[django,otel_django]"

# or pin the exact pre-release version
pip install "i-dot-ai-utilities[django,otel_django]==0.6.0rc1"

From Test PyPI (pre-release → test channel): point your installer at Test PyPI, provide production PyPI as a fallback index so runtime dependencies resolve, and pin the exact timestamped version the workflow generated (replace the version as required):

uv pip install --index https://test.pypi.org/simple/ --index-strategy unsafe-best-match "i-dot-ai-utilities[django,otel_django]==0.1.1rc202506301522"

Or with pip:

pip install --index-url https://test.pypi.org/simple/ --extra-index-url https://pypi.org/simple/ "i-dot-ai-utilities[django,otel_django]==0.1.1rc202506301522"

--pre alone is not enough for the Test PyPI channel: the package only exists on Test PyPI, so you must also point the installer at that index (and keep production PyPI as a fallback for dependencies).

Licence

MIT