agnes-media

In a DeepSeek Harness conversation, the agent used to be text-only. Wanting an illustration or a short clip meant leaving the chat and opening another tool. agnes-media registers Agnes AI image and video generation as native host tools, so the agent can produce them in-flow: describe a scene and it returns a 1024x1024 image or a 5-second 720P video, rendered inline in the DSH Web UI with no client-side plugin needed.

Three generation paths are covered. Text-to-image runs on agnes-image-2.5-flash with custom sizing and up to 4 concurrent outputs. Text-to-video runs on agnes-video-2.5-flash at 4 to 12 seconds across six aspect ratios including 16:9, 9:16, and 21:9. Image-to-video accepts one or up to five reference image URLs to turn static assets into motion. Transient 429 or 5xx errors trigger automatic backoff retries, and video polling caps at roughly six minutes, so the experience feels like submit-and-wait.

The plugin is deliberately zero-dependency, relying solely on built-in Node.js modules with no third-party packages to install or update. API calls are free, and mainland-Chinese users get a direct connection to apihub.agnes-ai.cn out of the box, no proxy required. Credentials can be set via environment variables or the DSH graphical credentials UI; the key is transmitted only as a Bearer header and is never persisted or logged.

Ideally suited to creators and developers who need a quick illustration, a storyboard clip, or a visual sanity check mid-conversation. In a mainland-China network setting, the combination of free access, no proxy, and zero configuration makes it the lightest media-generation entry point in the DSH ecosystem.