How hkuds/videoagent works
VideoAgent is a conversational, multi-agent toolkit for video understanding, editing, and creative remaking. The documented workflow starts with a natural-language request, maps intent to agent capabilities, and constructs an adaptive tool workflow; the runtime launcher hands control to the multi-agent framework. Major documented branches include Q&A and summarization, clip editing and beat synchronization, commentary and overview creation, and creative remakes such as memes, music videos, and cross-cultural comedy. The tree supports these role paths, and sampled VideoRAG code supports video ingestion and retrieval. Most role internals and dispatch details were not sampled, so those links reflect README-level documentation rather than inspected calls.
Main entry points: multi.py Multi-agent framework · videoragcontent.py Video RAG · vid_conversion.py Video conversion · main.py Application launcher
Components 22
Agent orchestration
- Application launcher
main.py - Multi-agent framework
environment/agents/multi.py - Agent base
environment/agents/base.py - LLM configuration
environment/config/llm.py
Video understanding
- Video Q&A
environment/roles/vid_qa/content_loader.py - Video summaries
environment/roles/vid_summ/summ_loader.py - Video preparation
environment/roles/vid_preloader.py - Visual search
environment/roles/vid_searcher.py
Editing and storytelling
- Clip editing
environment/roles/vid_editor.py - Beat detection
environment/roles/vid_rhythm/rhythm_detector.py - Beat-synced edits
environment/roles/vid_rhythm/rhythm_story_gen.py - Commentary video
environment/roles/vid_comm/comm_story_gen.py - News storytelling
environment/roles/vid_news/news_story_gen.py
Creative audio video
- Video conversion
environment/roles/vid_conversion.py - Cross-cultural comedy
environment/roles/cross_talk/cross_talk_adapter.py - Stand-up remake
environment/roles/stand_up/stand_up_adapter.py - Voice generation
environment/roles/voice_generator.py - Audio processing
environment/roles/mixer.py
Video retrieval
- Video RAG
tools/videorag/videoragcontent.py - Video segment index
tools/videorag/_storage/vdb_nanovectordb.py - Video metadata
tools/videorag/_storage/kv_json.py
Other components
- User
Connections 19
- User → Application launcher: starts conversation
- Application launcher → Multi-agent framework: starts framework
- Multi-agent framework → Agent base: uses agent base
- Multi-agent framework → Video Q&A: dispatches by intent
- Multi-agent framework → Video summaries: dispatches by intent
- Multi-agent framework → Clip editing: dispatches by intent
- Multi-agent framework → Beat-synced edits: dispatches by intent
- Multi-agent framework → Commentary video: dispatches by intent
- Multi-agent framework → News storytelling: dispatches by intent
- Multi-agent framework → Cross-cultural comedy: dispatches by intent
- Multi-agent framework → Stand-up remake: dispatches by intent
- Multi-agent framework → Video RAG: retrieves video context
- Video RAG → Video metadata: writes metadata
- Video RAG → Video segment index: indexes segments
- Video RAG → Video segment index: queries features
- Video preparation → Visual search: supports retrieval
- Beat-synced edits → Beat detection: uses beat analysis
- Video conversion → Voice generation: uses generated voice
- Video conversion → Audio processing: processes audio