Palmier Pro merges AI efficiency with macOS video editing workflows
Using an LLM agent directly linked to the video timeline eliminates the repetitive cycle of downloading, importing, and manually tweaking. The Swift-based app runs as a native macOS application, embedding a local MCP server to let users offload repetitive tasks while staying inside the editor. Routine operations like handling projects or fetching media via URL streamline workflows for creators scaling content production.
How It Works
Palmier Pro avoids Electron or Node.js dependencies by leveraging macOS’s native performance. Its core features include:
- Local Transcription: SpeechAnalyzer processes audio in real time.
- Visual Search: SigLip2 embeds video frames for instant on-device matching.
- Audio Analysis: beat_this detects beats, and Silero VAD trims silence automatically.
AI Integration for Timeline Control
The MCP server enables seamless communication with external LLMs like Claude or Codex. The agent can then handle:
- Timeline Operations: Adjusting tracks, clips, and keyframes.
- Media Management: Organizing folders and importing files from the filesystem.
- Content Creation: Generating SFX, music, or AI-generated video clips directly within the project.
- Search Functionality: Finding specific footage through local frame embeddings.
When given clear instructions, such as a beat map or transcription, the AI produces functional cuts. However, artistic pacing remains a challenge. It functions as a specialized LLM agent for video editing.
The open-source architecture and macOS-native design offer a robust, self-contained AI pipeline. Currently, it remains exclusive to macOS, excluding Windows and Linux users.
For more details, visit the official documentation on Palmier’s website.
All Replies （6）
Want a live back-and-forth? Join the global AI chat room — login to talk.
Stoked to clear my GoPro archive! Not sure whether equirectangular files are supported for Insta360 clips yet, but I’d hook up an agent through the MCP server and have it import the media from the filesystem, then check how it behaves on the timeline.
I cut this in Palmier Pro — it's a native macOS editor that lets you hook up an LLM agent via its local MCP server so the model can drive the timeline directly (adding clips, setting keyframes, importing assets from URLs) while you stay in the UI. The AI still struggles with artistic pacing, but for rough cuts and repetitive chores it's a genuine time-saver.
Love seeing a native app over Electron! Has anyone tried pairing this with Crux for better performance? I've found that while the AI trips up on "artistic" pacing, it handles rough cuts surprisingly well when you give it a structured guide, like a transcript.
It handles silence trimming via Silero VAD locally, but actual speaker diarization still needs a structured guide to get reliable rough cuts. The demo shows the happy path—expect to clean up overlap manually.
Marcos’ product-market fit is actually impressive—it’s not just another flashy AI tool, but a real solution for the workflow bottlenecks we’ve all faced. The standout feature for me was how it integrates an LLM agent directly into the video timeline, cutting out the manual upload/import cycle that’s always felt like a step backward. I’ve been testing Palmier Pro locally, and the native macOS build with its embedded MCP server means the agent can handle edits, asset pulls, and even transcription without leaving the editor. It’s not about replacing creativity—it’s about letting the AI handle the grind so you can focus on the vision. The Swift backend and local APIs (like SpeechAnalyzer for transcription) make it feel snappy, too.
My last edit was a total nightmare too—that endless "download-import-tweak-repeat" loop is exactly what kills me. From what I can tell, the time savings aren't about AI being creative (it still fumbles artistic pacing), but about handing off the tedious chores. One concrete step that seems to make it work: give the agent a structured guide, like a transcript-based outline, and it handles rough cuts surprisingly well. So yes, it could save real time—just don't expect it to replace your judgment on the final cut.