Korvai

A video editor driven by your own Claude Code

Korvai is our own native macOS video editor with a real multi-track timeline that you can trim, split and arrange by hand, plus a built-in bridge that lets your own Claude Code make and revise edits from a chat panel. A brand kit of logos, colors, fonts, music and caption style keeps the agent's edits on-brand.

  • 2026
  • macOS (Swift)
  • Claude Code bridge
  • MCP tool server
  • Metal compositor
  • On-device transcription
Illustration of Korvai

Why we’re building it

On-brand video edits usually mean one of two things: doing every cut by hand, or handing the whole job to a fully automated tool and losing control over the result. We wanted a real timeline that stays editable by hand at all times, with an agent that can also drive that same timeline directly through chat, so a request like a 45-second vertical cut with captions and a logo becomes an editable starting point instead of a black box.

Who uses it

Solo creator
Chats a cut request to Claude Code, then trims, retimes or swaps clips by hand on the real timeline.
Brand owner
Sets up a brand kit once, logo, colors, fonts, music, tone notes, so every video and every agent edit stays consistent.

What we built

A Swift app with a Metal compositor, a multi-track timeline (trim, split, ripple delete, crossfades) and brand kits for logos, colors, fonts, music and captions. Claude Code reads and edits the timeline through a built-in MCP tool server, with one undo step per agent turn and approval for anything outside its tools. Transcription runs on device for captions and silence removal.

  • Multi-track timeline

    Video, audio and text tracks with trim, split, ripple delete, drag, snapping, zoom and J/K/L playback.

  • Claude Code chat bridge

    The user's own claude CLI drives the timeline directly through an in-app MCP server, with streamed replies and tool-call cards.

  • One undo step per agent turn

    Every agent turn is a single checkpoint; Keep or Undo turn reverts it in one action without touching manual edits.

  • Brand kits

    Logo, color palette, fonts, music tracks, intro/outro clips, caption style and tone notes, generated into a project CLAUDE.md context file.

  • On-device transcription

    An on-device speech model produces word timestamps locally for burned-in captions and filler-word or silence-based cuts.

  • Approval cards

    Anything the agent wants to do outside Korvai's own editing tools, shell commands, file writes, web access, needs explicit approval.

  • Crossfades and transforms

    Position, scale, crop and crossfade transitions for reframing between 16:9, 9:16 and 1:1 formats.

How it works

  1. Create a brand: logo, colors, fonts, music, intro/outro and one editing style.
  2. Start a video under that brand: pick a format, write a brief, add reference images and footage.
  3. Ask in chat for a cut, for example a 45-second vertical edit in the brand's fast-shorts style with captions.
  4. Watch Claude Code build the cut on the real timeline through the in-app MCP tools.
  5. Fix anything by hand, or undo the whole agent turn in one step if it went wrong.
  6. Export an MP4 in H.264 or HEVC, at 1080p or 4K.
Close-up of Korvai's timeline with an agent-proposed clip highlighted beside the Claude Code chat panel.
Close-up of Korvai's timeline with an agent-proposed clip highlighted beside the Claude Code chat panel.

Where the AI fits

Korvai launches the user's own Claude Code CLI as a subprocess and talks to it over stream-json, so there is no Venzpire-hosted model or account involved. An in-app MCP server exposes editor tools, read the timeline, see frame grids, make specific edits, so the agent works on the real project rather than generating a separate file. An on-device speech model runs for transcripts used in captions and silence or filler-word removal. Every agent turn is a single undoable checkpoint, and anything outside Korvai's own tool set needs explicit user approval before it runs.

  • Bridges to the user's own Claude Code account over stream-json, no hosted inference by Venzpire
  • In-app MCP tool server exposes read/see/edit access to the real timeline
  • On-device transcription for captions and silence/filler-word cuts
  • One undo step per agent turn; approval cards gate anything outside editing tools

Every part of the product

  • macOS app

    Swift + Metal

    The native editor: timeline, brand kit editor, agent chat panel and export, all in one app.

  • Timeline editor

    Metal compositor

    Real-time preview with layered video, audio and text tracks edited by hand or by the agent.

  • Brand kit editor

    Swift

    Where logo, colors, fonts, music and caption style are set up and generated into project context files.

  • Agent chat panel

    Claude Code bridge + MCP

    Streamed chat with the user's own claude CLI, tool-call cards, inline frame grids and approval cards.

How it’s built

Korvai's compositor is Metal-based for real-time layered preview, sitting under a standard non-destructive timeline model (commands, transactions, snapshot undo) that both the user and the agent write to through the same code path, so a manual trim and an agent-driven trim are indistinguishable to the undo system except for turn grouping. The Claude Code bridge starts the user's own claude CLI per video session with stream-json in and out, narrowed flags to keep the user's own ~/.claude config and plugins from leaking into a Korvai session while still allowing normal account sign-in, and an in-app MCP server over a local bearer-token HTTP endpoint for the editor tool calls. Brand kits generate a project CLAUDE.md so brand tone, do/don't notes and style choices are the context Claude reads rather than being re-explained every time.

  • Swift
  • Metal
  • AVFoundation
  • Claude Code CLI
  • MCP
  • On-device transcription
  • stream-json

Related services: AI agent development

What clients say about working with us

Verified Fiverr reviews from our clients, quoted word for word.

5.0 98 verified reviews · Fiverr Top Rated
Excellent work. We have completed several projects with Venzpire_biz, and he is our trusted developer. We will continue working with him on further Android app projects.
projectsmdmRepeat buyerSpain · AI Websites & Software
I am very happy with everything Venkey did for me. I will certainly work with him in the future
alexsimion2019Repeat buyerFrance · AI Websites & Software
The team understands the requirement impressively well and they've built a truly complex and working product for me. They are always up to date with the AI trends and applied immediately in my product too. I'm very much satisfied with their work and highly recommended for any AI projects with Web app and mobile apps.
dnk3201Repeat buyerIndia · AI Websites & Software

Read all 98 client reviews

Questions about Korvai

What does Korvai do?

Korvai is a native macOS video editor from Venzpire with a real multi-track timeline, plus a built-in bridge that lets a user's own Claude Code make and revise edits in chat. Brand kits of logo, colors, fonts, music and caption style keep both manual and agent-driven edits on-brand.

Does Korvai use its own AI model to edit video?

No. Korvai bridges to the user's own Claude Code account over stream-json and an in-app MCP server, so the agent editing a project is the user's own Claude Code, not a model Venzpire hosts. Transcription for captions runs on-device with an on-device speech model.

Can I undo an edit Claude Code made in Korvai?

Yes. Venzpire built Korvai so every agent turn counts as one undo step, with a Keep or Undo turn choice shown right in the chat, separate from the manual undo history on the timeline.

Want a product built with the same care?

Tell us what you are building. Venktesh reads every application himself and replies within one business day.

Apply for a strategy call