OPENThe Edit Agent
A chat panel that operates the editor through a typed tool surface, and asks approval before it writes.
The Edit Agent is not a copilot that writes suggestions for you to apply by hand. It calls the same operations the interface calls, through a typed tool surface in the Electron main process.
The approval model
- Read-only calls run immediately — reading the timeline, inspecting assets, sampling frames
- Anything that writes to your project or starts a job pauses for approval — placing or trimming clips, generation jobs, exports
- There is no unattended mode. A conversation cannot silently modify a project while you are away
It can see the footage
Ask about a clip and frames are sampled and handed to the model as images with timestamps, so the answer comes from what is actually on screen rather than from a file name. This needs a vision-capable model; a text-only local model will still edit, but cannot watch.
Sessions
Conversations are kept per project as sessions. Start a new one, switch back to an earlier one, or delete it. History is stored in a path-free chats.json inside the project folder, and the start page lists past conversations so picking one reopens its project with the transcript restored.
The boundary it works inside
The agent reaches your project only through the typed tool surface. It has no shell access, and the renderer never handles filesystem paths, FFmpeg arguments, API keys, or OAuth tokens. See Agent tools for the full list and Data and privacy for the boundary.
Choosing its model
The agent's model comes from Settings → Edit Agent, and the chat panel has its own picker. Only providers you have connected offer models — including none but a local Ollama engine. See Providers and models.