About this project
Director is a framework for creating AI agents that work with video through natural-language instructions. It is positioned as a chat-based assistant for media workflows and is built on VideoDB's infrastructure for video storage, indexing, and streaming.
The project includes more than 20 pre-built agents for tasks such as summarizing videos, searching for specific moments, creating highlights and clips, organizing media libraries, generating movies from scripts, dubbing or editing audio and video, translating content, and adding subtitles. Agents can also add overlays, extract frames, compile content, and connect to external LLMs, databases, and generative AI APIs.
The main components are a backend reasoning engine, a chat-based user interface, a video player, and a collection view for browsing media. The reasoning engine interprets user requests, maintains context, selects and coordinates multiple agents, breaks work into steps, and reports progress. Example workflows include uploading a video, detecting scenes, creating a clip, and sharing the result to Slack.
Director can be run locally or deployed to a cloud environment. The documented setup requires Python 3.9+, Node.js 22.8.0+, npm, and API keys configured through environment files. A setup script prepares the frontend and backend environments, while Make commands can run both services together or separately. macOS and Linux are supported, along with Windows through WSL. One-click deployment references for Render and Railway are also provided.
Developers can add custom agents by duplicating the sample agent, renaming the class and metadata, implementing the run logic, defining supported content types, publishing progress and final responses, handling errors, and registering the agent in the chat handler. The repository includes local documentation served with MkDocs and accepts community contributions and new agent integrations.
Comments
0 Rating appears after 10 ratings
Sign in to join the discussion.