54xkeee/dsh-youreyes
Vision toolkit for text-only DeepSeek: model-invokable vision tool, wrapper adapters for deepseek/opencode-go (v4 flash/pro), Antigravity IDE quota (default, flash/pro) / any OpenAI-compatible VLM / Gemini / local Ollama channels, evidence memory with compaction rehydration, content-hash cache, and a bilingual client panel.
'Eyes for text-only DeepSeek': paste images, screenshots, or file paths into DeepSeek Harness and the model can answer image-related questions — DeepSeek stays the brain, vision is just the eyes. Images become text placeholders automatically. Key features: Antigravity IDE quota by default (flash/pro tiers) or free local Ollama with zero config; vision evidence memory (results persist in the session, reused across turns, restored after compaction); content-hash cache (same image + question recognized once per process); auto detail escalation for complex scenes; and a winCurl fallback for WSL/firewalled networks.
Install
dsh plugin --profile web add dsh-youreyesnpm package dsh-youreyes 0.1.0 (registry-verified 2026-08-28): dsh plugin --profile web add dsh-youreyes. Node.js >=20. Default vision is Antigravity IDE quota (flash/pro tiers) when the IDE is running, with free local Ollama auto-detection otherwise; any OpenAI-compatible endpoint, Gemini, or local Ollama work with your own key. Vision evidence persists in the session, and a content-hash cache means the same image + question is recognized once per process.
Compatibility
DeepSeek Harness. Node.js >=20. Text-only DeepSeek models gain image understanding via Antigravity / OpenAI-compatible / Gemini / local Ollama.
Details
- Repo: 54xkeee/dsh-youreyes
- Category: Vision & Multimodal
- Stars: 2
- Version: npm dsh-youreyes 0.1.0 (registry-verified 2026-08-28)
- Last push: 2026-08-18
- First seen: 2026-08-15
Recent updates
paste-to-see vision; Antigravity/OpenAI-compatible/Gemini/Ollama; vision evidence memory; content-hash cache; auto detail escalation; winCurl fallback.
FAQ
- Which vision providers?
- Antigravity by default (IDE quota when running), plus any OpenAI-compatible endpoint, Gemini, or local Ollama — your key just works.
- Does it remember what it saw?
- Yes — vision evidence persists in the session, is reused across turns, and is restored after compaction; a content-hash cache avoids re-recognizing the same image.
- Does it cost extra?
- It uses Antigravity IDE quota by default when the IDE is running, and free local Ollama otherwise — no per-image registration needed.
Alternatives
54xkeee/dsh-vision · Einskyle/dsh-llm-vision-bridge · Flyvhidbwo/dsh-vision-proxy