anyai

The unified gateway to the any* AI ecosystem — every AI task, one import.

anyai is the umbrella meta-package for the any* ecosystem. It provides a single import and a unified one-liner API for computer vision, OCR, LLMs, NLP, tabular ML, and deployment. The core package ships with zero heavy dependencies and includes useful rule-based implementations out of the box, while every sister package (anycv, anyocr, anyllm, anynlp, anyml, tableai, traincv, anydeploy, anyrobo) can be unlocked as an optional extra.

Built by Viet-Anh Nguyen at NRL.ai.

Why anyai?

• One-liner API — Get started in 3 lines of code for any AI task
• Plugin architecture — Discovers and delegates to installed any* sister packages automatically
• Local-first — Built-in rule-based methods work offline with no downloads
• Minimal core deps — Only pillow and pyyaml required; heavy ML is optional
• Production-ready — Type hints, tests, dataclass result types, graceful degradation

Installation

pip install anyai

For optional features:

pip install anyai[cv]        # + anycv (object detection, classification)
pip install anyai[ocr]       # + anyocr (text extraction from images)
pip install anyai[llm]       # + anyllm (LLM abstraction layer)
pip install anyai[nlp]       # + anynlp (NER, sentiment, summarization)
pip install anyai[ml]        # + anyml (AutoML for tabular data)
pip install anyai[table]     # + tableai (DataFrame profiling & cleaning)
pip install anyai[deploy]    # + anydeploy (ONNX/TFLite export + serving)
pip install anyai[robo]      # + anyrobo (voice agent framework)
pip install anyai[all]       # everything

Python 3.8+ supported (tested on 3.8, 3.9, 3.10, 3.11, 3.12, 3.13)

Quick Start

import anyai

# 1. Text summarization (built-in, zero deps — extractive sentence scoring)
summary = anyai.summarize(
    "Long article text here...",
    max_sentences=3,
)
print(summary)

# 2. Sentiment analysis (built-in AFINN-style lexicon with negation handling)
sentiment = anyai.sentiment("I absolutely love this library!")
print(sentiment.label, sentiment.score)   # "positive" 0.87

# 3. Keyword extraction (built-in TF-IDF-like scoring)
keywords = anyai.keywords("Machine learning is transforming software.", top_k=5)

# 4. Image metadata (built-in via Pillow)
info = anyai.image_info("photo.jpg")
print(info.width, info.height, info.format)

# 5. Delegated tasks (require sister packages)
labels = anyai.detect("photo.jpg")        # needs anyai[cv]
text   = anyai.ocr("scan.png")            # needs anyai[ocr]
reply  = anyai.chat("Explain RAG in 1 line.")  # needs anyai[llm]

Models & Methods

Built-in (zero-dependency) implementations

Delegated backends (via optional extras)

When you call a task that requires a sister package, anyai dynamically imports the registered backend:

• anyai.detect / classify / segment -> anycv (YOLOv8 / MobileNetV2 / DeepLabV3 via ONNX Runtime)
• anyai.ocr -> anyocr (Surya / EasyOCR / PaddleOCR / Tesseract / Vision-LLM)
• anyai.chat / embed / tools -> anyllm (Ollama / llama.cpp / OpenAI / Anthropic / HF)
• anyai.ner / classify_text -> anynlp
• anyai.automl -> anyml (sklearn + XGBoost/LightGBM)
• anyai.profile_df / clean_df -> tableai
• anyai.export / serve -> anydeploy

API Reference

CLI Usage

anyai summarize article.txt --sentences 5
anyai sentiment "I really enjoyed the film"
anyai keywords document.txt --top 10
anyai info photo.jpg
anyai version

Examples

Build a pipeline that spans packages

from anyai import Pipeline

# Each step delegates to the right sister package (if installed)
pipe = Pipeline([
    ("ocr",       {"backend": "auto"}),    # anyocr
    ("summarize", {"max_sentences": 3}),   # built-in
    ("sentiment", {}),                     # built-in
])

result = pipe.run("scanned_report.png")
print(result["summary"], result["sentiment"])

Use config-driven defaults

# anyai.yaml
llm:
  provider: ollama
  model: llama3.1:8b
cv:
  model: yolov8n

import anyai

# All subsequent calls inherit these defaults
anyai.Config.load("anyai.yaml")
anyai.chat("Hello")   # routed to ollama/llama3.1:8b

Graceful degradation

import anyai

# If anyllm is not installed, fall back to the built-in extractive summary
try:
    summary = anyai.summarize_llm(long_text)   # abstractive (needs anyai[llm])
except anyai.BackendNotAvailable:
    summary = anyai.summarize(long_text)       # extractive fallback (built-in)

License

MIT (c) Viet-Anh Nguyen