# towhee-io/towhee

Towhee is a framework that is dedicated to making neural data processing pipelines simple and fast.

Repository: https://github.com/towhee-io/towhee
Canonical: https://ross.abutalabs.com/products/towhee
Homepage: https://towhee.io
Language: Python
License: Apache-2.0
License Family: permissive
Topics: machine-learning, convolutional-networks, embedding-vectors, embeddings, computer-vision, image-processing, video-processing, feature-extraction, image-retrieval, unstructured-data, feature-vector, transformer, milvus, towhee, vision-transformer, vit, pipeline, llm
Last push: 2024-10-18T00:01:12+00:00

## Health v2 (maintenance only)
Score: 23/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 0, release rhythm 8, longevity 100
- inputs: {"age_days": 1877, "days_push": 685, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: none
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 3454, forks 259 (observed 2026-08-28T04:08:05.397550+00:00)

## What it is
Towhee is a Python framework for building ETL pipelines that process unstructured data (images, video, text, audio) into embeddings using state-of-the-art machine learning models. It offers a Pythonic API for prototyping pipelines with 700+ pre-trained models and integrates with vector databases like Milvus for production use.

## Use cases
- generate embeddings from images for image retrieval
- encode text documents into vectors for semantic search
- build a RAG pipeline with local LLMs and knowledge retrieval
- extract features from video clips with deep learning models
- process unstructured data and load it into a vector database
- prototype ML pipelines in Python and optimize them for production
- run CLIP or BERT models to embed multimodal data

## When to choose
- you need to embed unstructured data (images, video, text, audio) into vectors for search or retrieval
- you want pre-built pipelines with SOTA models instead of wiring up torch/transformers yourself
- you are building a vector search or RAG application with Milvus or similar vector databases
- you want a Pythonic API to prototype ML pipelines and optimize them for production

## When to avoid
- you need a single specific model with full control over inference details rather than a pipeline framework
- you only do tabular or classical ML workloads with no unstructured data
- you need a framework with very active recent development and a large community
- your pipeline is simple enough that calling a model library directly is easier

## Facets
- artifact type: framework
- maturity: maintenance
- function: machine-learning, etl, rag, llm-inference, image-processing, video-processing, nlp, audio-processing, data-science
- domain: machine-learning, deep-learning, artificial-intelligence, computer-vision, large-language-models
- platform: python, cross-platform
- tags: embeddings, unstructured-data, vector-embeddings, pipeline-orchestration, feature-extraction, milvus, multimodal, pretrained-models, natural-language-processing, data-engineering, retrieval-augmented-generation, docker

## Member repositories
- towhee-io/towhee (main) score 23

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:08:05.397550+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-29T18:37:32.047241+00:00, confidence not recorded.
  - readme: https://github.com/towhee-io/towhee (fetched 2026-08-28T04:08:05.397550+00:00, sha 56ccf1bf621a)
  - homepage: https://towhee.io (fetched 2026-08-29T09:31:17.814742+00:00, sha b15d757fbd15)
  - site_page: https://docs.towhee.io/ (fetched 2026-08-29T09:31:17.823798+00:00, sha 5c4b7df1644e)
  - registry_pypi: https://pypi.org/pypi/towhee/json (fetched 2026-08-29T09:31:17.826184+00:00, sha be29331ba377)
- Data as of 2026-08-30T08:39:29.467469+00:00.