Hand it an episode and it gives you back the viral moments cropped 9:16 with captions burned in, ready to drop into TikTok, YouTube Shorts, or Reels. The whole thing runs from one binary on your laptop, and you can drive it from Claude Code by saying “clip this episode.”
Once it's installed, the only thing that needs the internet is the optional AI scoring step (and only if you bring an API key for it).
Word-level timestamps and speaker diarization. Cached by file hash, so re-runs are instant.
YuNet face detection plus per-clip mouth-motion speaker tracking. No diarization needed for 2-person split-screen. Scene-cut guard kills jittery pans.
4 styles (branded, hormozi, karaoke, subtle), rendered via Remotion. Synced to word timing, filler stripped.
Drop a video, review clips, get a live 9:16 phone preview with optional TikTok wireframe, export. All in-browser.
Claude or Codex scores moments on four dimensions against your knowledge base. Checks the episode database so you don't ship the same clip twice.
Add to Claude Code, Claude Desktop, or Codex. Tell it "clip this episode" and it handles the rest.
Auto-picks VideoToolbox on Mac, NVENC on NVIDIA, VAAPI on Linux. CPU fallback works too.
13 markdown files you fill out once. Teach the AI your voice, title formulas, and banned words.
CLI, Web UI, or AI agent. Same pipeline.
Whisper + pyannote for word timestamps and speaker labels
Claude or Codex reads the transcript plus your knowledge base and picks clips
Batch render 1080×1920 MP4s with captions and −14 LUFS audio
8 options per clip, descriptions with hooks and SEO keywords
Two-line text briefs for podcast (16:9) and shorts (9:16)
Pre-upload checklist, first-24-hours ops, day 3 to 4 optimization
01-03: video engine. 04-06: PodStack slash commands.
Burned in at word-level timing. Filler words stripped.
--caption-style brandedDark box on the active word. Optional logo. The safe default.
--caption-style hormoziBold uppercase with a yellow highlight. For clips that need to hit hard.
--caption-style karaokeWhole line is visible. Words light up as they're spoken. Works for story clips.
--caption-style subtleSmall clean text at the bottom. Stays out of the way.
Nine slash commands that work in Claude Code, Codex, Cursor, or any AI coding tool that reads markdown. Plan the episode before you record. After, turn the transcript into titles, descriptions, and thumbnails you can paste straight into YouTube.
/plan-episodeBefore you record: drafts questions, story arc, and moments to steer toward
/process-transcriptPulls 8 to 15 moments and scores each one on hook strength
/generate-titlesWrites 8 title options per clip and checks them against your rules
/generate-descriptionsWrites descriptions with hooks, hashtags, and SEO keywords you can paste in
/plan-thumbnailsTwo-line thumbnail briefs for the 16:9 and 9:16 formats
/review-contentFix-first review. Parallel subagents catch voice, banned words, weak hooks, SEO, titles
/produce-shortsRuns the whole pipeline from transcript to a ready-to-publish bundle
/publish-checklistSteps for pre-upload, launch day, first 24 hours, and day 3 to 4
/retro-episodeLooks at how the episode did, then appends patterns to knowledge/13-learnings.md
Add it to Claude Code, Claude Desktop, or Codex. Say "clip this episode" and it runs transcription through export.
See all 19 MCP tools// claude_desktop_config.json or codex setup { "mcpServers": { "podcli": { "command": "node", "args": ["dist/index.js"], "cwd": "./podcli" } } }
Real pricing, what each one actually does, and what changes when you move.
See the open-source landscapeOpusClip is hosted, polished, and zero-setup. Podcli is the open-source local version: no watermark, no minute caps, your files never leave your laptop.
Vizard is a hosted clipper with a clean dashboard. Podcli is the local-first open-source version: no monthly cap, no watermark, no upload, scriptable from the CLI.
Submagic does captions on clips you already have. Podcli does the full pipeline: finds the clip, crops it, captions it, exports it. Captions are open-source React components you can edit.
ClipsAI is a Python library that gives you clip-detection and dynamic 9:16 reframing primitives. Podcli is the end-to-end product around the same idea: CLI, web UI, captions, AI scoring, MCP server.
Both are open-source. SupoClip's path leads to a hosted upsell. Podcli stays local-first, ships an MCP server, and adds a knowledge-base layer so the AI picks clips that sound like your show.
Descript is a full transcript-driven editor for whole episodes. Podcli is the clip-only pipeline that runs after the episode is edited: it finds the moments, crops 9:16, burns captions, and exports.
Riverside is where you record the podcast. Podcli is where you clip it. The overlap is Riverside's Magic Clips feature, and that's the part Podcli does better, free, and locally.
Podcli is open source (AGPL-3.0) and runs on your laptop. There's no watermark on the output, no monthly minute cap, and no signup. Your podcast files stay on your machine. Hosted SaaS clippers charge a subscription and require you to upload the source video, which is a different tradeoff for a different audience.
No. Transcription, clip picking, cropping, and export all run on your machine. The only network calls are to the Claude or Codex API when you ask for AI clip suggestions, and even those are optional.
Yes. That's the job. Podcli takes a long-form podcast episode, picks the moments worth clipping, crops to 1080×1920 with face tracking, burns in captions, and gives you files ready to upload to TikTok, YouTube Shorts, and Reels.
ClipsAI is a Python library you wire into your own pipeline. SupoClip is an open-source clipper with a hosted upsell. Podcli ships three entry points (CLI, web UI, MCP server) and includes a knowledge base layer, so Claude or Codex picks clips that match your show's voice, banned words, and title formulas rather than picking on transcript heuristics alone.
macOS, Linux, and Windows. Hardware encoding uses VideoToolbox on Mac, NVENC on NVIDIA GPUs, and VAAPI on Linux, with a CPU fallback for anything else.
Podcli sends the transcript and your knowledge base to Claude or Codex. The model returns scored clip candidates with timestamps. The knowledge base is a folder of markdown files that teaches the model your voice, title formulas, and banned words. Each candidate is scored on four dimensions, and the episode database is checked so you don't ship the same clip twice.
Podcli ships as a Model Context Protocol server with 19 tools. Plug it into Claude Code, Claude Desktop, or Codex and tell the agent to clip an episode. It handles transcription, clip scoring, and export end to end.
Claude Code, OpenAI Codex, Cursor, or any AI coding tool that reads markdown commands. setup.sh auto-detects your tool from existing folders and falls back to a generic install for everything else.
Yes. AGPL-3.0 licensed. Source is on GitHub. A commercial license is available if you need to use Podcli without AGPL terms — email siradze@nikusha.me.
CLI, Web UI, or AI agent. Same pipeline.