Voyager wants to become the Codex of creative work

Voyager allows models like Claude or ChatGPT to control Photoshop, After Effects, or Blender. A new approach to creative production.

The team behind Moda introduces Voyager, an agentic environment designed for creative work rather than software development. The principle is reminiscent of the harnesses used around Claude Code or Codex: a model is provided with tools, memory, files, and an environment in which to act. Voyager applies this logic to video, design, sound, 3D, and gaming.

The goal is to enable models like Claude, ChatGPT, Kimi, or DeepSeek to no longer just explain how to perform a creative task, but to intervene directly in the relevant software. Voyager can work with its own built-in engines or control already installed applications such as Photoshop, Illustrator, After Effects, Premiere Pro, DaVinci Resolve, Final Cut Pro, Blender, Unity, or Ableton Live.

The positioning is quite clear: where the first agentic harnesses were primarily built to write code, Voyager attempts to build the equivalent for making an edit, an animation, a musical composition, or a 3D scene. A harness rather than a new model

Voyager is not a generative model.

The platform acts as an orchestration layer around existing models. It provides them with a set of tools, specialized skills, memory, and the ability to observe or modify creative projects.

In its documentation, the team defines an agent harness as the software that surrounds a model and gives it access to the files, tools, and context necessary to act. Claude Code and Codex are cited as programming-oriented examples. Voyager adopts this principle with skills dedicated to editing, design, motion design, audio, and visual generation.

The model used can therefore change depending on the task or user preferences. Voyager notably works with existing Claude and ChatGPT connections as well as various external APIs.

This architecture also allows for the use of multiple models within a single project. For example, an agent can work on the storyboard, then call upon a video model like Seedance, H3 Max, or Veo, and then use editing software to assemble or correct the results. Voyager directly controls creative software

The most technically interesting part concerns how Voyager communicates with existing software.

The team favors native scripting interfaces when they exist: Photoshop scripts, After Effects and Illustrator scripts, DaVinci Resolve APIs, or a local bridge for Premiere Pro. When certain commands are not exposed by these interfaces, the agent can complete its action through direct control of the application, subject to permissions granted on macOS.

The principle therefore differs from a simple chatbot that merely knows a software's documentation.

Voyager can open a project, modify elements, apply settings, and then leave behind a standard file that remains editable in the original application.

For Blender, the team explains, for example, that they combine skills indicating to the agent which operations to perform with tools capable of executing those actions on the local scene. The agent can build a scene, run a render, or generate previews without replacing the existing Blender file.

In After Effects, this approach relies notably on the possibilities offered by ExtendScript and the various bridges available around the application. Three built-in engines to work without Adobe or Ableton

Voyager does not, however, depend entirely on third-party software.

The application includes three in-house engines: Voyager Design, Voyager Video, and Voyager DAW. They are integrated for free into the application and run locally, with no watermarks or limits on the number of projects, according to the company.

Voyager Design combines vector layers, raster, painting, layout, presentations, and keyframe-based motion design. It can export to PNG, SVG, PDF, PSD, Lottie, or ProRes 4444 with transparency.

Voyager Video features a timeline, multicamera tools, color functions, and keying. Projects can be transferred to Resolve or Premiere via OTIO, FCPXML, or EDL, and then exported in formats including H.264, HEVC, ProRes, or DNxHR.

Voyager DAW, for its part, covers arrangement, instruments, VST3 plugins, mixing, mastering, stems, MIDI export, and picture-synced composition. The team indicates that 132 of the 162 core features they compare to Ableton Live and Logic Pro are already supported, with another 27 partially available. This metric, however, remains internal to Voyager. An agent that can compose without a music generator

The audio component is a major part of the project.

In her launch demonstration, Anvisha Pai shows Claude Opus 5.5 composing a music track inside Voyager without using a music generation model. Instead, the agent works with the composition and arrangement tools available in the DAW.

This distinction is interesting: Voyager does not solely serve as a gateway to models capable of directly producing a song or a video. The model can also use traditional tools, select instruments, modify a structure, or work with existing plugins.

The same principle applies to images and video: generation can occur as one step among others, not necessarily as the core of the workflow. Video generation becomes an orchestrated step

Voyager can also call upon specialized models when the task requires it.

The team notably highlights Seedance, H3 Max, and Veo to produce shots intended for commercials or cinematic sequences. Voyager can help build the storyboard, character sheets, or necessary references beforehand, then retrieve the renders and integrate them into an edit.

The content published by the team shows several workflows ranging from a simple character sketch to a finalized video sequence. In one example, the process begins with an illustration, goes through a character sheet and several visual studies before resulting in an animated shot.

Another workflow uses character replacement on an existing video while preserving the editing, titles, audio, and timing in the editable project.

The goal is therefore less about creating an additional video model and more about managing the transition between multiple systems.