# jd-opensource/JoyAI-VL-Interaction

JoyAI-VL-Interaction: An Open Real-time Video-Language Interaction System

Repository: https://github.com/jd-opensource/JoyAI-VL-Interaction
Canonical: https://ross.abutalabs.com/products/joyai-vl-interaction
Homepage: https://joyai-vl-video-future-academy-jd.github.io/JoyAI-VL-Interaction/
Language: Python
License: Apache-2.0
License Family: permissive
Last push: 2026-08-25T07:44:18+00:00

## Health v2 (maintenance only)
Score: 58/100 (v2, computed 2026-09-02T17:46:02.011165+00:00)
- activity 99, release rhythm 35, longevity 5
- inputs: {"age_days": 83, "days_push": 8, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases, young
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1806, forks 180 (observed 2026-08-28T04:05:39.140148+00:00)

## What it is
JoyAI-VL-Interaction is an open 8B-scale vision-language interaction model with a complete deployable real-time streaming system, including the model weights, training recipe, time-aligned interaction dataset, and vLLM-based inference stack. It watches live video from a webcam or livestream and proactively speaks up in real time (sub-second latency) when a moment matters, rather than waiting to be asked.

## Use cases
- build an always-present AI assistant that watches a live camera feed
- real-time video call assistant that reacts without being polled
- monitor a livestream and get proactive commentary on important moments
- self-host a vision-language model for streaming video understanding
- train or fine-tune a proactive video-language interaction model
- deploy a quantized 8B VLM with vLLM on CUDA GPUs

## When to choose
- you need a fully open (weights, data, training recipe) real-time video-language model
- you want proactive, time-aware interaction rather than turn-based Q&A
- you need sub-second latency streaming inference on standard GPU infrastructure
- you want quantized checkpoints (INT4/INT8/FP8/NVFP4) for efficient deployment

## When to avoid
- you only need offline video analysis or batch video captioning
- you lack CUDA GPUs or cannot run 8B-scale models
- you need a turn-based chatbot rather than a proactive streaming assistant
- you need a small model for edge or CPU-only devices

## Facets
- artifact type: framework
- maturity: active
- function: llm-inference, machine-learning, video-processing, speech-recognition, streaming, rag
- domain: artificial-intelligence, large-language-models, computer-vision, deep-learning
- platform: python, cloud
- tags: vision-language-model, real-time-interaction, vllm, 8b-model, proactive-ai, streaming-video, self-hosted-ai, quantization, video, real-time, gpu, linux, docker

## Member repositories
- jd-opensource/JoyAI-VL-Interaction (main) score 58

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:05:39.140148+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 2026-08-30T03:21:29.906636+00:00, confidence not recorded.
  - readme: https://github.com/jd-opensource/JoyAI-VL-Interaction (fetched 2026-08-28T04:05:39.140148+00:00, sha 58c2392fb656)
  - homepage: https://joyai-vl-video-future-academy-jd.github.io/JoyAI-VL-Interaction/ (fetched 2026-08-29T11:00:38.341851+00:00, sha 39712938068e)
- Data as of 2026-08-30T08:39:29.467469+00:00.