Installation#

AutoIntent supports Python >=3.10,<3.15 (see requires-python in the package metadata).

Since v0.3.0, pip install autointent installs only the core runtime: PyTorch, scikit-learn, Optuna, vector search, datasets, and other always-on dependencies. Heavier or integration-specific libraries are shipped as optional extras.

Minimal install#

For the smallest footprint (core AutoML and modules that do not require optional stacks):

pip install autointent

If a preset or module needs an extra, install it with pip install "autointent[<extra>]" or combine several names in the brackets (comma-separated).

Optional extras matrix#

The table below lists each published extra, what it is typically used for in AutoIntent, and the matching pip command.

Extra	What it enables	Install command
`sentence-transformers`	SentenceTransformer embedders and pipelines that rely on them (for example many `classic-*` and embedding-based scorers).	`pip install "autointent[sentence-transformers]"`
`catboost`	`CatBoostScorer` and CatBoost-based tuning paths (see the modules API).	`pip install "autointent[catboost]"`
`transformers`	Hugging Face `transformers` models used by transformer presets and modules.	`pip install "autointent[transformers]"`
`peft`	Parameter-efficient fine-tuning (LoRA and similar) used together with transformer presets.	`pip install "autointent[peft]"`
`openai`	OpenAI API clients for OpenAI-backed embedders and the `zero-shot-llm` style workflows.	`pip install "autointent[openai]"`
`dspy`	DSPy-based augmentation and related generation utilities.	`pip install "autointent[dspy]"`
`wandb`	Weights & Biases experiment logging integration.	`pip install "autointent[wandb]"`
`codecarbon`	CodeCarbon energy / emissions tracking during runs.	`pip install "autointent[codecarbon]"`
`fastapi`	HTTP serving stack (FastAPI, Uvicorn, settings) for the AutoIntent server mode.	`pip install "autointent[fastapi]"`
`fastmcp`	FastMCP-based MCP server integration.	`pip install "autointent[fastmcp]"`
`opensearch`	OpenSearch client for OpenSearch-backed vector / retrieval integrations.	`pip install "autointent[opensearch]"`
`vllm`	vLLM as an optional high-throughput inference backend where supported.	`pip install "autointent[vllm]"`

Recommended bundles#

Combine extras in one install by listing them inside the brackets (no spaces inside the list).

Classic embedding + gradient boosting pipeline (typical classic-light / classic-heavy style workflows with SentenceTransformers and CatBoost):

pip install "autointent[sentence-transformers,catboost]"

Transformer presets (transformers-* presets and fine-tuning): install both Hugging Face transformers and PEFT:

pip install "autointent[transformers,peft]"

Zero-shot LLM preset (OpenAI API):

pip install "autointent[openai]"

HTTP API server:

pip install "autointent[fastapi]"

MCP server (FastMCP):

pip install "autointent[fastmcp]"

Experiment tracking (pick what you use):

pip install "autointent[wandb]"
pip install "autointent[codecarbon]"

`codecarbon` vs `fastmcp`#

codecarbon and fastmcp are declared as conflicting extras. A single environment should not select both at once; pick the extra that matches your goal (emissions tracking or FastMCP), or use separate environments if you genuinely need both stacks isolated.

Development install from Git#

Clone the upstream repository and use uv with the project Makefile (see CONTRIBUTING.md in the repository root for details):

git clone https://github.com/deeppavlov/AutoIntent.git
cd AutoIntent
make install

This installs the full contributor dependency set (tests, typing, docs, and so on), not only the minimal PyPI autointent wheel dependencies.