binsarjr/dsh-codex-media

README excerpt

dsh-codex-media Image and document analysis tools for agent harnesses, powered by a local OpenAI Codex CLI — with zero runtime dependencies (Node 22+ built-ins only). Two tools, one engine: - analyze image — describe a local image or answer a question about it (PNG, JPEG, WebP, GIF). The image is attached natively via codex exec --image . - analyze document — analyze a local document (PDF, Office, RTF, or text formats) or answer a focused question about it. The model reads the file and returns only the answer, so binary documents never flood the agent's context with extracted text. - generate image — generate an image from a text prompt and save it to a local file. Defaults to a Hermes Agent oneshot (authenticated through your existing ChatGPT/Codex login — no API key), with the OpenAI Images API as the deterministic alternative. Why this exists Text-only models (like DeepSeek) cannot receive images through a model's native attachment channel — DeepSeek Harness rejects the message with "The current model does not support images" . The robust pattern is to deliver file paths instead, and let a separate tool analyze the file. dsh-codex-media is the analysis half of that pattern: it o…

Category