Tesseract gives AI agents an editing engine instead of an interface to click through

Mirage is opening Tesseract, a video environment designed to be controlled directly by agents such as ChatGPT. Editing, keyframes, compositions, motion design, and audio remain structured inside an editable project, with previews and rendering handled locally on Mac or Windows.

The software interface moves behind the agent

Having an agent make precise video changes usually means letting it operate software designed for humans or rebuilding the result through code. Tesseract takes a different route: the agent gets direct access to the creative engine and can modify the project structure without navigating Premiere or After Effects.

The project still retains familiar professional video concepts. Keyframes, compositions, adjustment layers, timing, effects, typography, and sound remain native and editable elements. Multiple clips, for example, can keep their own framing, effects, and timing inside the same composition.

The main difference is the control layer. A natural-language instruction can shorten a shot, move a title, adjust a keyframe, or rebalance audio without rebuilding the rest of the project. ChatGPT becomes the editing interface

With the Tesseract plugin for ChatGPT, users provide footage, images, screen recordings, audio, and a brief. The agent can then build the project, generate a preview, receive revisions, and render the finished video.

Changes can remain local to individual elements. A title can be revised without touching the edit, a single effect parameter can change independently of the rest of a scene, and multiple audio tracks can remain separate. Tesseract therefore preserves an editable state between instructions instead of producing only a flattened video after every request.

Mirage also distinguishes this architecture from video tools built around HTML or React. The agent manipulates editing and motion-design concepts directly instead of representing the video as a webpage or web application. A local engine behind Captions

The engine runs locally on Apple Silicon and Intel Macs as well as 64-bit Windows 10 or later. Linux, WSL, and cloud execution are not supported. A browser-only ChatGPT conversation therefore cannot run Tesseract by itself: the agent needs a local environment with access to files and execution tools.

The local engine is free and does not require a Mirage account, while the agent's own subscription or model usage remains separate. Tesseract can also be installed as skills inside other compatible local agent environments.

The underlying technology is not starting from scratch. Mirage says the same engine also powers Captions, its video creation application. Tesseract now exposes that infrastructure directly to agents and developers. No footage generation

Tesseract does not generate footage or AI avatars. It works with existing videos, images, and audio, whether recorded or previously created with other models. Its scope covers editing, compositing, motion graphics, and sound.

The engine can also sit behind a custom interface. An agent can build manual controls for timing, typography, color, or framing, then continue working on the same editable project after a user makes a manual adjustment.

Tesseract's distinction is therefore less about automating clicks inside existing software and more about removing that intermediary layer: the agent and the user work on the same structured video project, while interacting with it through different controls.