Aito Python SDK

:Info: Check Github <https://github.com/AitoDotAI/aito-python-tools>_ for the latest source code and documentation <https://aito-python-sdk.readthedocs.io/en/latest/>__ for more information. :Maintainer: Aito [email protected]

|PyPI| |PyPI version|

About

The Aito Python SDK is an open-source library that helps you to integrate your Python application to Aito <https://aito.ai/>_ quicker and more efficiently.

The SDK also includes the Aito Command Line Interface (CLI) <https://aito-python-sdk.readthedocs.io/en/latest/cli.html>_ that enables you to interact with Aito using commands in your command-line shell, e.g: infer a table schema from a file or upload a file to Aito.

API versions

Each Aito API version has its own package, whose meaning never changes:

.. code-block:: python

from aito.v2 import Client      # the v2 API  (`from aito.v1 import Client` for v1)

# the public read-only sandbox; use your own instance URL and key in production
client = Client('https://shared.aito.ai/db/aito-demo',
                'yg4rTlXkqDzm4y8gPeY75HCKaNwfbTQ2si64ONTi', env='v2')
prediction = client.predict(
    from_table='invoices', where={'Description': 'cloud services'}, predict='GLCode')
print(prediction.first.value, prediction.first.probability)   # E002 0.83...

aito.Client is the default version, and moves only on a major release of this package: since aitoai 1.0 it is the v2 client (0.x: v1). Quickstarts can use it; production code should import the version it means, aito.v1 or aito.v2, so an upgrade cannot change the API underneath it. To stay on v1: from aito.v1 import Client, or pin aitoai<1.

The pre-0.7 paths (aito.client, aito.client.v2, aito.api) were removed in 1.0; importing one raises an ImportError naming its replacement. The command-line tool still talks to the v1 API in 1.x: v2 support arrives as an opt-in in a 1.x minor release, and the CLI's default moves to v2 only in 2.0. The design is written up in docs/versioned-namespaces.md.

pip install aitoai installs the API clients. The command-line tool, schema inference and file conversion need the dataframe toolchain: pip install 'aitoai[cli]'.

Errors and retries

A v2 error is an AitoV2Error with the server's code and status_code; branch on those, not on the message. is_not_found is a missing table or collection.

The client repeats a request only when that is safe, waiting the server's Retry-After plus jitter:

• 409 write.contention: a write that lost to concurrent writes on the same table and committed nothing. Up to 3 retries (write_contention_retries=; 0 turns it off). Other 409s are never repeated.
• 429: the server was busy and turned the request away. Up to 3 retries, and at most 60 s of waiting for one request (rate_limit_retries=; 0 turns it off).

It does not repeat a job's status or result (a stored result does not change: submit the job again), or a 503: a write answered 503 may or may not have been applied, so read before repeating it. With the retries off, or after the last one, err.is_retryable says whether repeating the same request later is safe:

.. code-block:: python

from aito.v2 import Client, AitoV2Error

client = Client(url, key, rate_limit_retries=0)   # this caller queues its own retries
try:
    prediction = client.predict(from_table='invoices', where=invoice, predict='GLCode')
except AitoV2Error as err:
    if not err.is_retryable:
        raise
    queue_for_later(invoice)

Transient failures (a 502, 503 or 504, a dropped connection, a timeout) are retried only if you opt in, and only for requests that change nothing: reads, queries and optimize, never a write: a batch insert sent twice stores its rows twice. The exception is an upsert into a collection with a primary key, create_collection(name, columns, primary_key='id') and then upload_entries(name, rows, on_conflict='update'): it lands once however often it's sent, so it's retried too. Client(url, key, retries=3, backoff=0.5) waits up to backoff * 2**n seconds before retry n. The default is retries=0. What each client does on each failure, and how to resume a failed upload: docs/client-resilience.md.

Run Aito locally

With Docker installed, aito start runs a local Aito and stores its keys, so the client needs no arguments:

.. code-block:: console

$ pip install aitoai
$ aito start          # pulls a pinned image, listens on http://127.0.0.1:9005, prints the keys
$ python -c "import aito; print(aito.Client().get_schema())"

aito stop stops it (data and keys are kept), aito status checks it, aito keys prints the keys again (--rotate replaces them), aito logs shows its log and aito upgrade moves to the engine version this package pins. These work without the [cli] extra. From Python, aito.Client() with no arguments uses AITO_URL and AITO_API_KEY if set, else the local server's profile.

examples/v2_quickstart.py is a runnable end-to-end example that creates a collection, loads it, predicts, explains, evaluates and drops it again.

Use Aito from an AI assistant (MCP)

pip install 'aitoai[mcp]' (Python 3.10+) adds aito-mcp, an MCP server for coding agents and assistants. Its tools are predict, recommend, relate, match, search, query, evaluate and schema/upload. Each tool's description says when Aito fits and when another tool (a language model, a trained model, a search engine) is the better choice. Results carry $p and a flattened $why. The server finds the instance the same way as aito.Client(). It is read-only unless started with AITO_MCP_ALLOW_WRITES=1.

.. code-block:: console

$ claude mcp add aito -- aito-mcp      # Claude Code; any MCP client runs `aito-mcp` over stdio

Setup for Claude Code and other assistants: Use Aito from Claude Code and other AI assistants <https://aito.ai/docs/articles/use-aito-from-claude-code-and-other-ai-assistants/>. When Aito fits and when not: When to use Aito, and when not <https://aito.ai/docs/articles/when-to-use-aito-and-when-not/>.

claude-plugin/ is a Claude Code plugin with this server and a skill, add-predictive-feature. The skill walks an agent through a fit check, a schema from an existing table, a first prediction, a threshold chosen with _evaluate, and act / assist / abstain wiring.

Support / Feedback

For issues with, questions about, or feedback, please join our support channels <https://aito.ai/join-slack/>__.

Installation

Aito Python SDK can be installed with pip <http://pypi.python.org/pypi/pip>_

.. code-block:: console

$ pip install aitoai

Check our installation guide <https://aito-python-sdk.readthedocs.io/en/latest/install.html>_ for more information

Documentation

You will need sphinx <https://www.sphinx-doc.org/en/master/>_ installed to generate the documentation.

$ pip install sphinx
$ pip install sphinx_rtd_theme

Documentation can be generated by:

.. code-block:: console

$ cd docs
$ make clean html

Generated documentation can be found in the docs/build/html/* directory.

Dependencies:

The Python SDK SQL integration and the tests depend on the ODBC db drivers and the postgres and mysql functionality.

The main drivers can be installed in Debian based Linuxes with:

sudo apt install unixodbc-dev odbc-postgresql postgresql-client default-mysql-client
pip install pyodbc

You can find more information about installing the MySQL ODBC drivers here

Also note: that the database integration tests require local postgres / mySQL servers.

Environment:

To run the test, you need to have the following environment variables defined:

• AITO_GROCERY_DEMO_INSTANCE_URL
• AITO_GROCERY_DEMO_API_KEY
• AITO_INSTANCE_URL
• AITO_API_KEY

Testing

Install the required packages:

.. code-block:: console

$ pip install -r requirements/build.txt
$ pip install -r requirements/test.txt

You can use our test cli to run tests:

.. code-block:: console

$ python -m tests -h

.. note::

Some tests require some environment variables to be setup. Use the test CLI list command to discover and display the test suites and cases

The live tests run against a local Aito loaded with a fixture they own (tests/fixtures/grocery), as CI does. Run one with your own test keys, load the fixture, and point the variables at it:

.. code-block:: console

$ docker run -d -p 127.0.0.1:9005:9005 -e READ_WRITE_APIKEY=<rw-test-key> -e APIKEY=<ro-test-key> \
    ghcr.io/aitohq/aito:v2.12.0
$ export AITO_INSTANCE_URL=http://localhost:9005 AITO_API_KEY=<rw-test-key>
$ export AITO_GROCERY_DEMO_INSTANCE_URL=$AITO_INSTANCE_URL AITO_GROCERY_DEMO_API_KEY=<rw-test-key>
$ python tests/fixtures/grocery/seed.py

The seed replaces the database's schema, so it refuses any host but localhost. A local engine has no S3, so the file-upload tests take the batch fallback there; the S3 path itself runs against a cloud instance in CI's release-branch smoke. AITO_SKIP_LIVE_TESTS=1 skips the live tests when no instance is set, as the Windows CI job does.

There are multiple test suites:

1. CLI suite:

.. code-block:: console

$ python -m tests -v suite cli

.. note::

You need an Aito instance and set up the credentials with ```AITO_INSTANCE_URL``` and ```AITO_API_KEY``` to test some functions.

2. SDK suite:

.. code-block:: console

$ python -m tests -v suite sdk

.. note::

You need an Aito instance and set up the credentials with ```AITO_INSTANCE_URL``` and ```AITO_API_KEY``` to test some functions

3. SQL functions tests:

• Test against Postgres:

  .. code-block:: console

  $ python -m tests -v case sql_functions.test_connection.TestPostgresConnection
  $ python -m tests -v case sql_functions.test_cli_sql_functions.TestPostgresFunctions
  

• Test against MySQL:

  .. code-block:: console

  $ python -m tests -v case sql_functions.test_connection.TestMySQLConnection $ python -m tests -v case sql_functions.test_cli_sql_functions.TestMySQLFunctions

.. note::

To test the SQL functions, you need to install `pyodbc <https://pypi.org/project/pyodbc/>`_ and the specific database ODBC driver.

4. Build and test built package:

• To build the package:

  .. code-block:: console

  $ pip install -r requirements/deploy.txt
  $ ./scripts/deploy test.pypi --no-publish -b dev
  

  WARNING: This will update the version in 'aito/init' file. DO NOT commit this to Git!

• The built wheel should be at dist/. To install:

  .. code-block:: console

  $ pip install dist/aitoai-<version>.whl
  

• To test the built package, create an environment variable TEST_BUILT_PACKAGE and run the above tests

5. Test the documentation:

• To test the inline documentation:

  .. code-block:: console

  $ python -m tests -v suite inline_docs

• To test the code blocks in rst files:

  .. code-block:: console

  $ cd docs $ curl -o reddit_sample.csv https://raw.githubusercontent.com/AitoDotAI/kickstart/master/reddit_sample.csv $ export SPHINX_DISABLE_MOCK_REQUIREMENTS=TRUE $ make doctest

Contributing

1. Making changes
2. Add unittest to the appropriate test suite (e.g: test case for Aito Schema: tests/sdk/test_aito_schema.py) or create a new test suite inside the tests folder
3. Add documentation:

• Inline documentation if applicable
• Update .rst file in docs/source folder (e.g: Add a new Client function to docs/source/sdk.rst)

4. Add inline doc test if applicable (e.g: inline doc test for Aito Client: tests/inline_docs/test_client_inline_docs.py)

.. note::

Remember to update the autodoc_mock_imports in conf.py file if there are additional requirements

5. Update the version in aito/init to e.g. '1.2.3rc1' and write release notes in docs/changelog.rst
6. Check CircleCI and issue a PR
7. Deploy to production (scripts/deploy) with the appropriate version after the PR is reviewed

.. |PyPI| image:: https://img.shields.io/pypi/v/aitoai?style=plastic :target: https://pypi.org/project/aitoai/ .. |PyPI version| image:: https://img.shields.io/pypi/pyversions/aitoai?style=plastic :target: https://github.com/AitoDotAI/aito-python-tools