Arsaze is an AI-native video editor built for both human editors and AI agents. It presents a real, multi-track timeline — described by its makers as "Cursor for video editing" — where users can cut, rewrite, generate, grade and review footage by hand, or hand control to an AI agent that drives the same timeline. The product positions a single workspace for AI video editing, AI video generation and agent-driven video editing, aimed at people who want a genuine non-linear editor rather than a one-click automation toy.
The product addresses a workflow shaped by fragmentation and feedback loss. Generation tools typically live in separate subscriptions, so material has to be downloaded and re-imported before it can be cut, and review notes left by clients tend not to survive the next cut, arriving detached from the frame they describe. Arsaze's answer is to put generation, grading, review and editing on one timeline, and to make that timeline addressable by an AI agent. Its pitch to that audience is simple: stop scrubbing, start directing. Rather than replacing the editing suite with a single automated button, it keeps a conventional multi-track editor at the center and adds an agent as an operator.
The first of Arsaze's four power tools is Rewrite, which lets editors treat video like a document. Every clip is transcribed word by word, so the transcript becomes an editable surface: select a sentence and delete it, and the corresponding cut happens on the timeline. The same transcript-driven approach is used to find retakes, strip filler words and shorten pauses. Because the timeline follows the transcript, cleaning up a talking-head recording becomes a text-editing task rather than a scrubbing task. Arsaze also provides word-timed automatic captions in styles that can be reused across a project, and captions are editable like any other clip on the timeline.
The Playground is a node editor for generation, usable by both the editor and their agents. Users chain prompts and AI image and video models into reusable graphs — the site gives examples such as Gemini Image into Seedance, or two stills into a single transition — and then drop the result straight onto the timeline. Beyond the Playground, generation is built into the editor across three media types with one credit balance on every paid plan, so there is no tab-hopping between five subscriptions and no downloading and re-importing files. AI video generation covers text-to-video and image-to-video from frontier models including Veo 3.1, Seedance 2.5, Kling 3.0 and PixVerse v6, landing on the timeline as real clips. AI image generation covers stills, thumbnails, style frames and b-roll plates with Nano Banana Pro, Flux 2 Pro, Seedream and Gemini Image. AI audio adds natural voiceover in dozens of languages plus music and sound effects, with ElevenLabs v3 and Multilingual v2 listed.
Color is Arsaze's full grading stack, and it can be driven by hand or by asking. It includes a keyer, shot matching, a light tool, color wheels, curves, HSL qualifiers, film-stock LUTs and finishing. Grading is non-destructive and applied per shot, then rendered on export, and editors can import their own .cube files. The site highlights wheels, curves and HSL controls, film and creative LUTs, and the ability to match one shot's look to another. On the audio side, Arsaze offers voice isolation, noise reduction, loudness normalization and automatic ducking under speech. These tools matter because every one of them is also a tool the agent can call, so a request to grade a sequence teal and orange, or to clean a noisy talking-head recording, plays out on the real timeline.
Arsaze runs on a dual engine: a real-time multi-track timeline and a code-to-video renderer, side by side in the same edit. The timeline engine offers unlimited video and audio tracks with frame-accurate trims, keyframes, transitions and live preview — the NLE editors already know. The code-to-video engine renders title cards, kinetic type and animated scenes written as code, by a human or an agent, frame-perfect onto the timeline; the site shows a scene definition with a stagger of words, a blur range and an easing curve, rendered at 1080p. Rounding out the toolkit are Remarker, Arsaze's frame-accurate client review tool, automatic version history that versions every edit so cuts can be branched, compared and reverted, and export options covering MP4 from the cloud, free local exports on Windows, and FCPXML for DaVinci Resolve and Premiere Pro.
Arsaze works by treating the AI agent as an operator that drives the same timeline a human uses. Setup begins by connecting an agent: Arsaze is added as a custom connector in Claude, ChatGPT or Grok over MCP, a process the site says takes about two minutes and requires no code. The user then states what they want in natural language — the example given is "Cut the silences, add b-roll where I mention the car, grade it teal and orange" — and the agent edits the real timeline, with its tool calls playing back live in the editor. From there the editor reviews and tweaks, stepping in manually at any point, shares a review link, then exports MP4 or FCPXML. Crucially, the agent is optional: Arsaze is a full manual NLE with timeline, trim modes, color wheels and keyframes, and the agent is an operator you can hand the controls to rather than a requirement.
Benefits center on control and continuity. Because every edit is versioned automatically as it happens, nothing an agent does is one-way: if the agent makes a bad edit, the editor can step back to any earlier state. Feedback survives the next cut because Remarker pins client comments to the exact frame and version they concern, and because those notes are agent-readable, the agent can read every note and make the fix. Generation arriving directly on the timeline removes the download-and-reimport loop. Exports render on Arsaze's servers by default, so the editor is just a browser tab and no fast machine is required, while Windows users can export locally for free with no watermark. Uploads, generated video and voice models stay tied to the account and are not used to train models without consent.
Concrete use cases follow directly from these tools. A creator editing a talking-head video can run one pass to remove dead air and filler words, then fine-tune on the timeline and caption the result with word-timed captions. A marketer or filmmaker can ask an agent to cut silences, place b-roll where a topic is mentioned and grade the sequence teal and orange, then review the result and export. A motion designer can write title cards and kinetic type as code and render them frame-perfect onto the same timeline as the footage. A freelancer or studio can share a Remarker link so clients pin comments to exact frames, reply in threads and rate scenes. Teams finishing in DaVinci Resolve or Premiere Pro can export FCPXML with cuts, transitions and captions intact.
Arsaze is aimed at editors, creators, freelancers and studios who want AI assistance inside a real editing environment rather than a black-box generator, and at developers and agent users who already work in tools like Claude, ChatGPT, Grok, Codex, Gemini, Cursor, Windsurf, Zed, Copilot and Perplexity, whose logos the site displays as the agents it is built for. Officially, Claude, ChatGPT and Grok are supported today over MCP. Arsaze runs in the browser and offers a Windows desktop app. Pricing is freemium: a free plan with 50 welcome AI credits, no credit card required, covers audio analysis of videos such as transcripts, captions and silence detection. AI generation of video, image, voice and music and the built-in AI assistant come with paid plans, which add monthly AI credits and cloud export hours; credit top-ups never expire and local Windows exports are always free.
Arsaze's primary value proposition is a single, real timeline that both humans and AI agents can drive — the place where AI video editing, AI video generation and agent-driven video editing converge. By combining a manual NLE, a node-based generation Playground, a full color and audio stack, code-to-video rendering, frame-accurate client review, automatic version history and open exports to DaVinci Resolve and Premiere Pro, it lets editors direct rather than scrub, and lets agents do real work on a project that stays fully reversible and under human control.