Configuring dsh-drawai
DeepSeek Harness loads dsh-drawai from the profile where it is installed. The README documents desktop and web profiles separately; after installation, the corresponding host must be fully restarted. The canvas stores files in the current workspace. New canvases default to .drawai. Opening an older .drawio file attempts migration to a same-directory .drawai while retaining the original.
Optional environment variables control downloaded media components:
| Component |
Default location |
Configuration |
| Video runtime |
~/.dsh/drawai-video-runtime/ |
DRAWAI_VIDEO_RUNTIME_DIR, DRAWAI_RUNTIME_REGISTRY |
| Assets and fonts |
~/.dsh/drawai-assets/v1/ |
DRAWAI_ASSETS_DIR, DRAWAI_ASSETS_URL |
| FFmpeg |
~/.dsh/drawai-ffmpeg/<platform>-<architecture>/ |
DRAWAI_FFMPEG_PATH, DRAWAI_FFMPEG_DIR, DRAWAI_FFMPEG_MIRROR |
| Browser |
Available Chrome or Edge |
DRAWAI_CHROME_PATH |
Commands provided by dsh-drawai
The agent-facing diagram_read tool reads diagram structure, styles, selection and file revision. diagram_apply edits nodes and edges, and accepts expectRevision to prevent writes based on a stale file. diagram_export exports SVG or PNG from the host, including for channels without a canvas sidebar.
Video workflows add shotcraft_search, motion_project, motion_preview, motion_audio, motion_render and motion_voice; motion_render apply places a completed render back into a video node. motion_audio_verify checks the resulting audio track.
Requirements and behaviour
DeepSeek Harness must provide the plugin host and resolve @deepseek-ai/dsh-tools for the documented development checks. Basic canvas use does not require the media components, but first use of video, audio or some web capture paths may download a runtime, assets, FFmpeg or Chromium. FFmpeg is searched by explicit path, system PATH, cache and automatic download.
Known limits of dsh-drawai
The interface is currently Simplified Chinese. Existing pages can be edited, but new pages cannot be created in the interface. .drawio.svg, .png and .html embedded containers are unsupported; external or relative image paths do not display directly. Export supports SVG and PNG, not PDF. Audio nodes enter a rendered video only after being placed on its timeline, and voice generation depends on an available local synthesis service or system speech facility.
Written from the project's own documentation and kept in sync with it. Where the two disagree, the source is authoritative — read the README on GitHub