dsh-multi-tools

However strong a text-only model is at inference, creative work demands visual feedback. An indie game developer needs real-time Unity scene state, a CAD engineer wants to inspect a DXF render, a Ren'Py author needs to validate a storyboard screenshot, a video editor expects to see the Premiere timeline. Until now these tools live in separate silos and visual information shuttles through a human eye. dsh-multi-tools bundles the ModLens visual-evidence layer with nine creative-tool integrations into a single install, extending DeepSeek Harness from pure text inference into a perceptible, actionable creative workflow.

The core is ModLens: it turns image files into structured evidence for text-only models, backed by local Codex, an OpenAI-compatible endpoint, or local Ollama, all over loopback with zero cloud calls. On top of that, ComfyUI, Unity, After Effects, Photoshop, Premiere Pro, AutoCAD, SolidWorks, and Ren'Py integrations each toggle independently; enabling one auto-generates a dedicated Agent Preset, disabling one cleans up with no residual routes. The settings page runs a passive dependency check on open, and a live check contacts only loopback services or short-lived local MCP probes without installing, downloading, or launching anything external.

Built for developers who already run inference on DeepSeek Harness and work daily inside creative toolchains - indie games, video post, engineering CAD, visual novels, or any context where a text model must see and drive creative-software output. The entire suite is private, install-once, and cloud-free; every visual route is explicitly configured by the user, never guessed from a model name.