liustack/modlens

ModLens is a plugin that gives text-only AI models the ability to read images. The flagship chat models from DeepSeek and GLM can only process text, so they cannot see or interpret pictures. ModLens solves this by acting as a vision engine that sits alongside these models. You paste an image directly into the chat, and the plugin converts it into structured text evidence the model can understand. There is no need to save the image to a file first or pass a file path manually. The plugin works across several coding assistant tools including Claude Code, Codex, OpenCode, and Pi. For DeepSeek Harness specifically, installation is a single command that adds a native tool called modlens_read_image. When you paste an image, ModLens sends it to a vision engine and returns a full transcription, layout regions in reading order, and lists of entities and relations. The model then quotes specifics from that structured output rather than guessing or inventing what the image contained. ModLens supports ten vision sources in total. Six are built-in providers: a free Gemini API key, any OpenAI-compatible endpoint, an Anthropic API key, the free Antigravity CLI, a signed-in Claude Code, and a signed-in Kimi Code. The other four come from reusing logins in local agent CLIs you may already have. If you configure more than one, they form a failover chain where fast API providers try first and slower CLI backups step in if needed. Every attempt is recorded so you can see which engine produced the result. Setup is designed to require minimal configuration. The installer checks what you already have on your machine and asks before reusing existing logins. If nothing is available, a free Gemini key from Google AI Studio is recommended, which brings each image read down to 5 to 10 seconds. Uninstalling is as simple as deleting a folder, with no hooks or proxy daemons left behind. The full README is longer than what was shown.