feat(models): add DeepSeek V4 Pro and V4 Flash via OpenRouter

[incogbyte]

Summary

Adds the current DeepSeek models to Shift, routed through the OpenRouter provider (following the existing Grok/Mercury pattern):

• deepseek/deepseek-v4-pro — 1M context, reasoning ✅
• deepseek/deepseek-v4-flash — 1M context, reasoning ✅

Both are enabled by default for agent and float usage.

Why via OpenRouter (not a native deepseek provider)

Caido's AIUpstreamProviderId only supports anthropic | google | openai | openrouter. The ModelProvider enum in packages/shared/src/models.ts is constrained with satisfies Record<string, AIUpstreamProviderId>, so a native deepseek provider would fail to compile and Caido couldn't route it. OpenRouter is already a supported provider and exposes DeepSeek, so this is the clean, idiomatic path — no shared-type or frontend changes needed.

Why V4 models only

Per the DeepSeek API docs, deepseek-chat and deepseek-reasoner are deprecated on 2026/07/24 (they map to the non-thinking/thinking modes of deepseek-v4-flash). Only the current V4 models are included to avoid shipping something that breaks within weeks.

Changes

• packages/backend/src/models/openrouter.ts:
  • Added DEEPSEEK_V4_PRO and DEEPSEEK_V4_FLASH to OpenRouterModelIds
  • Added the two model entries to openrouterModels (provider OpenRouter, contextWindow: 1_000_000, reasoning: true)
  • Added both to defaultOpenRouterModelsConfig with ["agent", "float"]

Validation

• ✅ openrouter.spec.ts + stores/models/update.spec.ts → 17/17 passing
• ✅ tsc --noEmit (backend) → no errors

How to use

1. Configure your OpenRouter API key in Caido AI settings.
2. DeepSeek V4 Pro / DeepSeek V4 Flash appear in Shift's model list, enabled for agent and float.

🤖 Generated with Claude Code

[bebiksior]

Thanks!