A transparent API proxy that intercepts images in your requests, runs them through a vision model, and replaces them with text descriptions — before the text-only LLM ever sees the request. Works with Claude Code, Codex CLI, and any OpenAI-compatible client.
Open DashboardYour coding agent sends a request with an image. Vision Bridge catches it at the HTTP layer.
The image runs through Gemini, GPT-4o, or your own vision provider. Text comes back.
The text description replaces the image. DeepSeek (or any text model) responds as if it could see.
DeepSeek V4 dominates coding benchmarks at 1/50th the cost — but it can't process images. Seven open-source tools bridge this gap. None of them are products. Vision Bridge is the opinionated one: zero config, transparent proxy, cache images so repeats in a conversation are free. Point your tool at it and images just work.