docs: explain native vision and vision proxy

[Vizards]

Summary

• document DeepSeek V4 Flash Vision Exp as a standalone experimental native-vision model
• clarify how Flash and Pro use Vision Proxy, including automatic selection and custom endpoint behavior
• update model descriptions, settings copy, walkthrough content, and screenshots in English and Simplified Chinese

Validation

• npm run lint
• npm run format:check
• npm run compile
• git diff --check

[Copilot]

🟡 Changes recommended

Several descriptions incorrectly promise that images go directly to DeepSeek even when a custom API endpoint is configured.

Once you've addressed the issues Copilot identified, you can request another Copilot review.

Pull request overview

Documents native vision for Flash Vision Exp and Vision Proxy behavior for Flash/Pro across English and Simplified Chinese content.

Changes:

• Adds Flash Vision Exp to model and onboarding documentation.
• Explains native versus proxied image handling.
• Updates localized UI and extension descriptions.

File summaries

File	Description
src/i18n.ts	Updates localized vision UI copy.
resources/walkthrough/show-models.nls.zh-cn.md	Adds Vision Exp to Chinese walkthrough.
resources/walkthrough/show-models.md	Adds Vision Exp to walkthrough.
resources/walkthrough/set-api-key.nls.zh-cn.md	Updates Chinese setup instructions.
resources/walkthrough/set-api-key.md	Updates setup instructions.
README.zh-cn.md	Documents vision behavior in Chinese.
README.md	Documents vision behavior in English.
package.nls.zh-cn.json	Updates Chinese setting descriptions.
package.nls.json	Updates English setting descriptions.
package.json	Revises extension marketplace description.

Review details

Suppressed comments (1)

src/i18n.ts:221

• deepseek-copilot.baseUrl can point to a proxy or compatible provider, and native image bytes are sent to that configured endpoint rather than necessarily to DeepSeek. Stating “directly to DeepSeek” misrepresents the data destination; describe the configured API endpoint instead.

		'Experimental DeepSeek V4 Flash vision model that sends user image attachments directly to DeepSeek for native image understanding.',

• Files reviewed: 10/12 changed files
• Comments generated: 3
• Review effort level: Balanced

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.