Agentic Video Production

EDBOT

EdBot agent art

AI-powered video editing pipeline. Natural language commands → real edits. Local inference, zero cloud dependency.

37
Tools
57+
Endpoints
6
Agents
95%
Local

What

An AI Agent That Edits Video

EdBot is an agentic video production pipeline built on local LLMs and GPU-accelerated tooling. It ingests raw footage, transcribes speech, detects chapters, scores editorial quality, catalogues everything into a searchable library, and assembles rough cuts — all driven by natural language commands. No cloud APIs. No subscriptions. One machine, one pipeline, full control.

How

Seven-Stage Pipeline

Raw footage flows through seven processing stages. Four are live and battle-tested. The pipeline extends as each stage comes online.

▶
Ingest
LIVE
→
◉
Transcribe
LIVE
→
◈
Analyze
LIVE
→
▣
Catalogue
LIVE
→
◧
Rough Cut
→
◫
Export
PLANNED
→
◎
Distribute
PLANNED

Why

Hours Become Minutes

Manual video editing means scrubbing through hours of footage, labeling clips by hand, scoring takes by instinct, and building timelines one cut at a time. EdBot replaces that with NLP commands and automated analysis.

Before — Manual Workflow

  • Scrub hours of raw footage
  • Manually label every clip
  • Guess which takes are best
  • Hand-build timelines cut by cut
  • Re-watch to find specific dialogue
  • Export and re-export for each platform

After — EdBot Pipeline

  • Drop footage → auto-transcribed in minutes
  • Every segment indexed and searchable
  • Chapters scored by editorial quality
  • NLP search: "find all mentions of Scrooge"
  • Rough cut assembled from top-scoring chapters
  • Multi-format export: 16:9, 9:16, 1:1, ProRes

Footage

From the Library

Real frames from the catalogued footage library — 27 chapters extracted, scored, and indexed.

Chapter 1Chapter 3Chapter 5Chapter 7Chapter 10Chapter 12Chapter 14Chapter 16Chapter 19Chapter 21Chapter 23Chapter 25

Features

System Capabilities

Each feature is a self-contained module in the pipeline. Live features have real data behind them. Planned features are in development.

Pipeline LIVE
Seven-stage processing pipeline from raw ingest through to distribution. Four stages operational.
7 stages · 4 live
→
Transcription LIVE
GPU-accelerated speech-to-text via faster-whisper. Full-text search across all indexed segments.
148 segments indexed
→
Chapter Detection LIVE
NLP-based chapter boundary detection with multi-weighted editorial scoring across four dimensions.
27 chapters · scores 3.0–7.3
→
Footage Library LIVE
Catalogued footage with searchable metadata, auto-extracted thumbnails, and sortable indices.
20 clips · 27 thumbnails
→
Dialogue Matching LIVE
Cross-reference spoken content across takes and angles for best-take selection and continuity.
175 match groups
→
Silence Detection LIVE
Automated dead air identification with timestamp-level gap mapping per clip.
Auto-gap identification
→
Agent System LIVE
Six specialized AI agents orchestrated by AriBot. Local LLM inference via Ollama with per-agent model routing.
6 agents · 11 recipes
→
Automated Rough Cut NEXT
Score-driven clip assembly. Auto-sequence building from highest-scoring chapters in the catalogue.
→
Resolve Integration PLANNED
DaVinci Resolve Studio API integration for programmatic timeline building and multi-format export.
→
Multi-Cam Sync PLANNED
Feed-ID scoped session management with auto-alignment of camera angles and audio sources.
→
Analytics Loop PLANNED
Post-publish performance data feeds back into the edit cycle. Engagement metrics inform the next rough cut.
→
YouTube Shorts PLANNED
Portrait reframe pipeline for 9:16 output. Automated YouTube Shorts export from long-form content.
→
Subtitle Automation PLANNED
Transcription-to-SRT generation with GPU-accelerated subtitle burn via FFmpeg h264_nvenc.
→