pirate-608/dsh-multi-tools--plugins-modlens ↗★ 2
@pirate-608/dsh-modlens
Text-first DSH vision bridge using Codex CLI, OpenAI-compatible endpoints, and Ollama
安装
npx -p @deepseek-ai/dsh dsh plugin --profile web add github:pirate-608/dsh-multi-tools#6ac433e02e836f77218737a67747a8c32f376021&path:plugins/modlens说明文档
阅读完整 README ↗@pirate-608/dsh-modlens
A text-first vision bridge for DeepSeek Harness, narrowly derived from ModLens 3.22.0. It provides three named routes in the default order codex → openai → ollama and never installs Gemini or Antigravity providers.
dsh plugin --profile web add @pirate-608/dsh-modlens
Configure routes in the profile patch. DSH reuse requires consent: true; this reuses a local CLI login but may still upload the image and spend remote account quota. OpenAI-compatible credentials are DSH credential references. Ollama uses its native loopback /api/chat API and never pulls a model automatically.
- id: pirate-modlens
name: '@pirate-608/dsh-modlens'
config:
routes:
codex: { type: codex-cli, consent: true }
openai:
type: openai-compatible
baseUrl: https://vision.example.com/v1
model: vision-model
credentialRef: OPENAI_VISION_API_KEY
structuredOutput: true
ollama:
type: ollama
baseUrl: http://127.0.0.1:11434
model: qwen3-vl
failover: [codex, openai, ollama]
The package does not synthesize a fake multimodal LLM route. Text-only paste is converted to a private opaque reference, modlens_read_image returns structured evidence, and that ordinary tool result is retained in the session log.