Get Started

Reeltern Engineering Handbook

A local, reversible video editing engine, AI Director composer, and 31-tool Model Context Protocol (MCP) stdio server powered cleanly by FFmpeg and PyAV.

Local-first security contract: All imports are copied into content-addressed SHA-256 snapshots scoped inside --media-root. Original footage is never modified in place, and zero shell commands or external cloud APIs are exposed over MCP.
macOS & Windows desktop launch note: Unsigned open-source builds trigger macOS Gatekeeper / Windows SmartScreen on first run. On macOS, copy Reeltern Studio.app into Applications, click Done, then open System Settings → Privacy & Security → Open Anyway (or install the wheel with Python and run reeltern studio).
Model Context Protocol

Connect via Local MCP

Unlike remote cloud MCP bridges, Reeltern runs directly as a local stdio server next to your footage. Choose your coding assistant below and paste the generated configuration without modifying global system state.

~/.gemini/config/mcp_config.json
Generated client entry (equivalent to reeltern mcp-config)

              

Five AI Director Profiles

Both reeltern compose and the MCP compose_video tool share these five calibrated profiles:

Style Preset & Fit Grade & Motion Audio & Transition Captions & Overlay
viral_reel reels · blur_fill vivid · zoom / pan slideleft · voice EQ + duck kinetic + progress bar
cinematic fhd · cover cinematic · zoom crossfade · -16 LUFS + duck clean captions
tutorial fhd · contain studio · steady wipeleft · voice cleanup EQ boxed + progress bar
documentary fhd · contain vintage · pan / zoom dissolve · warm ducking clean soft + burned
noir_trailer fhd · cover noir · dramatic zoom fadeblack · -16 LUFS master bold high contrast

Closed-Loop Inspect → Audit Protocol

Every serious agent session follows four verified steps instead of firing blind filter strings:

1. inspect_media_visuals(asset_id)  → Read PNG contact sheet + scene_cuts[]
2. compose_video(assets, style=...) → Build immutable reversible Plan v2
3. render_plan(plan_id, preview=1)  → Cache unchanged segments, mix -16 LUFS
4. audit_render(job_id)             → Inspect ebur128 + black frames → final 1080p

CLI Reference

Every workflow is mirrored across the standalone Desktop Studio app, CLI, Python, and MCP:

reeltern studio
reeltern demo --workspace demo-project
reeltern compose clip.mp4 --style viral_reel --music bgm.wav --preview --audit
reeltern audit JOB_ID --workspace reeltern-workspace
reeltern import-audio soundtrack.mp3 --workspace reeltern-workspace
reeltern inspect clip.mp4 --thumbs 6 --samples 32
reeltern card "Chapter One" --subtitle "Intro" --duration 2.0
reeltern transcript clip.mp4 captions.srt --workspace reeltern-workspace
reeltern agent TRANSCRIPT_ID "Keep key highlights" --model qwen2.5:3b --style documentary
reeltern mcp-config antigravity --workspace ws --media-root footage

Architecture, Security & Validation

Deep technical guides live directly alongside the codebase on GitHub: