getting the vision model to work in DSH

Just posting this here in case anyone else has been trying to get it to work. What I found - dsh has two deepseek routes. The one you're probably on by default is deepseek-official , and that one is text-only and hard-coded that way. The adapter declares input: ['text'] for every model id you throw at it. So the vision model won't work there, no matter what. I had to go through the pi-ai route instead, where each model declares what it actually accepts. If anyone has better ways of doing it I'm all ears. I added this to ~/.dsh/settings.yaml : llm-pi-ai: providers: deepseek-vision: api: openai-completions baseURL: api.deepseek.com apiKeyEnv: DEEPSEEK_API_KEY models: - id: deepseek-v4-flash-vision-exp input: [text, image] then point the default at it agent-default-model: provider: deepseek-vision model: deepseek-v4-flash-vision-exp reasoningEffort: off the critical line is input: [text, image]. The read_image tool (and the API proxy) refused to hand an image to a model that hasn't declared image input. And the model settings page in the GUI only lets you change id/name/context window - there's no field for input, so you have to write that into the yaml yourself. It should hot-reload, no restart. New sessions pick it up. If you've already got a session open it stays pinned to the old route until you switch it in the model picker. Here's my full ~/.dsh/settings.yaml as an example (I also use dsh for local models hence the ollama section): ui-onboarding: welcomeNoticeVersion: 2026-08-13.1 agent-default-model: provider: deepseek-official model: deepseek-v4-pro reasoningEffort: high agent-presets: default: code ui-theme: preference: dark llm-pi-ai: providers: ollama: displayName: Ollama api: openai-completions baseURL: localhost:11434/v1 headers: Authorization: Bearer ollama compat: supportsDeveloperRole: false maxTokensField: max_tokens models: - id: qwen3:8b name: Qwen3 8B contextWindow: 32768 maxTokens: 8192 - id: qwen3.8:27b-ridge name: qwen3.8:27b-ridge contextWindow: 65536 maxTokens: 8192 deepseek-vision: displayName: DeepSeek Vision api: openai-completions baseURL: api.deepseek.com apiKeyEnv: DEEPSEEK_API_KEY models: - id: deepseek-v4-flash-vision-exp name: DeepSeek-V4-Flash-Vision (Exp) input: [ text, image ] contextWindow: 1000000 maxTokens: 384000