Add DeepSeek V4 built-in models

[alwaysmavs]

Summary

Add DeepSeek V4 Flash and DeepSeek V4 Pro to Wanta's built-in model catalog using the official model IDs deepseek-v4-flash and deepseek-v4-pro.

User impact

Signed-in users can select both DeepSeek models directly from the built-in-model section without configuring a DeepSeek API key. Existing DeepSeek BYOK presets remain available under the separate custom-model section, so credential ownership stays explicit.

Both built-in models are intentionally text-only. Wanta does not advertise image or native PDF input for them, while tool calling and the low, high, and max reasoning levels remain available.

Root cause

The DeepSeek V4 entries were already present as custom-provider presets but were not registered in the current built-in catalog. Persisted selections using their former built-in IDs were consequently treated as stale and fell back to Auto.

Changes

• Register DeepSeek V4 Flash and Pro on the OOMOL OpenAI-compatible runtime with their official model IDs.
• Configure a 1M context window, Wanta's standard 256K input tier, and the default 32K output limit.
• Mark both models as text-only with tool calling and low/high/max reasoning variants.
• Preserve historical persisted selections for the restored built-in IDs.
• Add registry, persistence, capability, routing, and OpenCode configuration coverage.

Validation

• pnpm run lint
• pnpm run ts-check
• pnpm exec oxfmt --check electron/models/builtin.ts electron/models/builtin.test.ts electron/models/store.test.ts electron/agent/agent.test.ts
• pnpm test — 283 test files and 2,126 tests passed
• git diff --check

[coderabbitai]

Summary by CodeRabbit

• New Features

  • Added DeepSeek V4 Flash and Pro as built-in model options.
  • Added low, high, and maximum reasoning variants.
  • Supports one-million-token context windows and tool calls.
  • Models use official OOMOL routing and support text input only.

• Bug Fixes

  • Preserved DeepSeek V4 Flash selection when previously saved in model settings.

Walkthrough

The built-in registry now includes DeepSeek V4 Flash and Pro. Both models use OOMOL routing, official model IDs, one-million-token contexts, reasoning variants, tool-call support, and text-only capabilities. Tests validate registry metadata, token limits, built-in ID recognition, catalog contents, agent behavior, and persisted model selection.

🚥 Pre-merge checks | ✅ 2 | ❌ 2

❌ Failed checks (2 warnings)

Check name	Status	Explanation	Resolution
Title check	⚠️ Warning	The title describes the main change but does not use the required <type>(<scope>): <subject> format.	Rewrite the title in the required format, for example feat(models): add DeepSeek V4 built-in models.
Description check	⚠️ Warning	The description explains the change and validation, but it omits the required Safety and Compatibility section and its required confirmations.	Add the Safety and Compatibility section and address each required checkbox item.

✅ Passed checks (2 passed)

Check name	Status	Explanation
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

✨ Finishing Touches

✨ Simplify code

• Create PR with simplified code
• Commit simplified code in branch codex/add-deepseek-v4-builtins

---

_{Comment @coderabbitai help to get the list of available commands.}