Vatt

Early access

Vatt / Approach The edit stays yours

You drive.
AI assists.
Every cut
stays editable.

Vatt is not a generator that hands you a locked file. It is an editor that reads your footage, drafts cuts, and keeps every decision open — so you can work by hand, guide it with a prompt, or let it run a full pass, then change anything.

Try AgentOpen Timeline
01 / Your footage02 / Your decisions
NLECopilotAutopilot
Explore the approach

Involvement

Three Levels of
AI Involvement

Vatt does not lock you into a single mode. You decide how much help AI gives on each pass, and you can switch levels at any point in the edit.

01 / NLE

NLE: Cut Every Frame by Hand

Treat Vatt like a traditional non-linear editor. You drag clips onto the timeline, trim frames at the boundary, arrange picture-in-picture and split-screen layouts by hand, and keyframe effects exactly where you want them. Nothing is hidden behind a prompt box, and AI stays completely out of the way until you explicitly invite it in.

This is the level creators fall back to when a moment matters more than speed. A punchline that needs three extra frames, a reveal that has to land on a beat, a layout the model keeps getting slightly wrong — you fix it directly instead of describing it. Manual work is slower, and that is the point: the last ten percent of a reaction video is usually where the personality lives.

02 / Copilot

Copilot: AI Suggests, You Approve

Copilot watches the same footage you do and offers concrete suggestions: where the strongest reactions peak, which silences to cut, what the captions should read, and which layout fits the moment. Every suggestion arrives on the timeline as a normal editable clip, marked so you can see what came from AI and what came from you.

You stay the editor. Accept a rough cut and then nudge two boundaries, keep the captions but drop the effect, or reject a whole pass and ask for a different read of the same section. Because suggestions are applied as ordinary timeline objects rather than a rendered result, changing your mind later costs a drag, not a re-export.

03 / Autopilot

Autopilot: A Full First Cut in One Pass

Hand Vatt your source video and your reaction take, and Autopilot assembles a complete first cut. It analyses speech, silence, emotion, and events in the source material, then decides where to cut, when to go full-frame, when to drop into picture-in-picture, and where captions belong across the whole runtime.

What you get back is a draft, not a deliverable. The full timeline is there to inspect — every cut, layout change, and caption is a clip you can move, replace, or delete. Most creators use Autopilot to skip the mechanical first pass on a long recording, then spend their time on the handful of moments that decide whether the video performs.

Direction

Two Ways to Direct
the Same Timeline

Whether you prefer direct manipulation or natural-language direction, both interfaces write to the same editable timeline. You can switch between them within the same project.

GUI

GUI: Drag, Trim, and Keyframe Directly

The graphical interface is the editor you already know: a timeline with layered tracks, an inspector for the selected clip, and a viewer that updates as you scrub. Every cut, marker, layout frame, caption, and keyframe is visible as an object on screen, and anything visible can be selected and changed by hand.

Direct manipulation wins whenever precision beats description. Aligning a zoom to an exact frame, stacking three layers with different opacities, or shifting one caption fifty milliseconds earlier is faster to do with a mouse than to explain in a sentence. The GUI is also where you verify AI work, because it shows the structure instead of summarising it.

LUI

LUI: Direct the Edit in Plain Language

The language interface lets you direct the edit in plain sentences. Restructure a section, cut down to the strongest reactions, tighten the pacing of the first minute, switch a stretch to split-screen, add captions in your usual style — you describe the intent and Vatt writes the result onto the timeline for you to review.

Prompts scale in a way that clicking does not. One instruction can touch forty minutes of footage, and a selection plus a short phrase can rework a section without you scrubbing through it first. It is the fastest way to explore a different structure, because rejecting a prompt costs one undo instead of an hour of manual rebuilding.

Access

One Timeline,
Many Entry Points

The language layer is not tied to a single window. Talking to the Agent inside the app is the entry point available today; the rest are planned surfaces for the same instruction layer, so a script, an agent, or your own pipeline can direct an edit the way you would by hand.

Available

Chat in the App

Describe the edit to the Agent while you watch the timeline, then review and refine what it writes.

Coming soon

Command Line (CLI)

Script repeatable edits, run a batch of recordings in one pass, and slot rendering into an existing pipeline.

Coming soon

MCP Server

Let agents and IDEs such as Cursor or Claude Code instruct Vatt directly, using the same language layer as in-app chat.

Coming soon

API and SDK

Call Vatt from your own backend or automation so an edit can start from an upload, a webhook, or a schedule.

Exploring

Agent Skills

Freeze a recurring editing routine into a reusable instruction pack an agent can run the same way every time.

The combinations

Every Combination,
Mapped

The combinations that actually exist across AI involvement and interface. Six out of nine are real workflows in Vatt.

GUI and Conversation are the entry points shipping today. Planned CLI, MCP, and API access all run through the same Conversation column.

AI involvement by editing interface
AI involvementGUI (Manual)Conversation (LUI)Zero Input (Auto)
No AITool Editing— (doesn't exist)—
AI AssistedSelect + PromptPrompt Editing—
Fully AI—One-Line DraftAuto Editing

FAQ

Approach
Questions.

Q01Is Vatt a manual editor or an AI editor?

Both. You can edit every frame by hand, ask the Agent to draft a section with a prompt, or run a one-click auto pass. Every output lands on the same editable timeline, so you can mix all three in one project.

Q02What is the difference between Agent and Timeline?

The Agent is a conversation layer: you describe what you want and it writes clips, layouts, captions and effects onto the timeline. The Timeline is the direct manipulation layer: you drag, trim, and keyframe the same clips. They edit the same project, so switching between them does not lose work.

Q03Can I undo something the AI did without undoing my own edits?

Yes. AI passes are tracked separately from manual edits, so you can roll back a specific Agent pass while keeping the changes you made before and after it.

Q04Do I have to use prompts?

No. Prompts are one entry point. If you prefer, you can import your footage and treat Vatt like a traditional editor — using AI only when you want speed, such as running Audio Align or finding reaction peaks.

Q05What makes Vatt different from a one-click AI video generator?

Generators produce a finished file from a prompt. Vatt produces an editable project from your own recording. The AI drafts, ranks, and suggests; you remain the editor and can change any decision.

Q06When should I use manual editing versus an AI pass?

Use an AI pass when you want a fast rough cut, need to find peaks in long footage, or want to try a different structure. Use manual editing for final timing, unusual creative choices, and anything where you already know exactly what you want. Most real edits alternate between the two.

Q07Does Vatt have a CLI or an API?

Not yet. Today Vatt is a desktop app with a timeline and an in-app Agent. A command line entry point and an API are on the roadmap so scripted, batch, and automated edits can drive the same timeline.

Q08Can I use Vatt from Cursor or another AI agent?

Not yet, but that is the direction. An MCP server is planned so agents and IDEs such as Cursor or Claude Code can send editing instructions to Vatt directly, using the same language layer as the in-app Agent.

The edit stays yours

Bring the Footage.
Keep the Timeline.

Get an invite and see how NLE, Copilot, and Autopilot fit into the same editable project.

Get invite codeRead the changelog