dsh-agnes-media

The DeepSeek Harness LLM seam speaks only chat protocols, yet Agnes AI image and video generation live on separate REST endpoints that return files rather than conversations. dsh-agnes-media bridges that gap by exposing two model-facing tools: agnes_image for text-to-image, image-to-image, and multi-image composition, and agnes_video for asynchronous text-to-video, keyframe generation, and multimodal reference tasks. Agents can request visual content as a natural step within their workflow without switching to a standalone HTTP client.

The plugin handles the practical edge cases you would otherwise have to code by hand. Credentials are resolved per call through the harness credential seam and never stored locally. Artifacts are downloaded as local files so the ephemeral URLs Agnes returns do not expire silently. Polling backs off on 429 and 503 responses instead of reporting a rate limit as a generation failure. The video model defaults to the free tier so an unconfigured call cannot silently incur charges. File extensions are sniffed from the downloaded bytes, the output directory follows the calling session workspace, and absolute paths override it.

Best for developers who have already built agent pipelines on DeepSeek Harness and want their models to call image or video generation in-context, and for teams integrating the Agnes AI ecosystem into a unified harness workflow.