Show the Kimi logo for Kimi models in the model picker

[warp-agent-staging]

Summary

Kimi models rendered the generic Warp agent glyph in the model picker instead of a provider logo.

Kimi is served through Fireworks, and the server reports Fireworks-hosted models as LlmProvider::UNKNOWN. LLMProvider::icon() returns None for Unknown, so model_leading_icon fell through to Icon::Agent and the picker branded Kimi rows as Warp.

Adding a first-party Kimi provider end to end (GraphQL enum + server mapping + BYOK plumbing) is out of scope for a logo, so the icon is keyed off the model id instead. Every Kimi model the server ships is kimi-* (kimi-k25-fireworks, kimi-k26-fireworks, kimi-k27-code-fireworks, kimi-k3-fireworks), so a plain kimi- prefix check covers them and leaves the other Fireworks-hosted families (GLM, MiniMax, Qwen, DeepSeek) alone. The branch sits after the existing auto / custom-router / Bedrock / Gemini-Enterprise cases, so badge priority is unchanged.

The asset is the official Kimi brand mark in the bundled-icon convention (24×24 viewBox, single tintable path), so warpui_core::elements::Icon recolors it from the theme — same shape as #14769 (Grok/xAI logo).

Changes (+16 / -1 across 4 files):

• app/assets/bundled/svg/kimi.svg — new bundled asset.
• crates/warp_core/src/ui/icons.rs — Icon::KimiLogo variant + path mapping.
• app/src/ai/llms.rs — one is_kimi_model branch in model_leading_icon.
• app/src/terminal/view/ambient_agent/model_selector.rs — the ambient agent model list gets the same branch, so both pickers agree.

Verification

No regression test — the requester explicitly waived it for this change ("It definitely doesn't need tests"). The behavior is a two-line icon mapping keyed off a literal id prefix. The pre-existing icon tests in app/src/ai/llms_tests.rs are untouched and still pass (33/33 in ai::llms::tests).

Repository checks:

• ./script/format --check — clean.
• cargo clippy -p warp -p warp_core --all-targets --tests -- -D warnings and cargo clippy -p warp --features gui --all-targets --tests -- -D warnings — clean.

Visual verification against a local --features gui build, driven through computer use on a real display:

• kimi k2.6, kimi k2.7 code, kimi k3 all render the Kimi mark.
• glm 5.2, minimax 2.7, minimax 3, qwen 3.6 plus, qwen 3.7 plus, deepseek v4 pro all still render the agent glyph — none picked up the Kimi logo.
• claude, gpt-5.x, gemini, grok icons unchanged; no blank or missing icons.

These screenshots were captured before the diff was shrunk and still stand: the rework only collapsed how the Kimi model is recognised, and the same rows resolve to the same icons, so the rendered result is byte-for-byte the same.

Rework changes

Requester feedback: "I thought this should be a three-liner. It definitely doesn't need tests". Diff reduced from +127/-3 across 5 files to +16/-1 across 4 files.

• Removed all three added tests from app/src/ai/llms_tests.rs (85 lines) per the explicit waiver above. That file is now absent from the diff entirely.
• Replaced the model_provider_icon / is_kimi_model / starts_with_kimi_token trio and its multi-field name scan with a single is_kimi_model one-liner matching the kimi- id prefix, plus one branch in model_leading_icon. Doc comment cut to two lines.
  • Tradeoff worth a reviewer's eye: this drops the display-name and base-model-name fallbacks, so a hypothetical future Kimi model whose id does not start with kimi- would not pick up the logo. Every id in the current catalogue does.
• Kept kimi.svg and the icons.rs mapping as-is.
• The ambient-agent selector change stayed at 2 lines plus the import, so both pickers remain consistent.
• Visual proof not re-captured: the rendered outcome is unchanged.

CHANGELOG-IMPROVEMENT: Kimi models now show the Kimi logo in the model picker.
Originating thread: https://warpdev.slack.com/archives/C0BDQDW8V5E/p1786075084680309

Computer-use screenshots (4)

Model picker filtered to 'kimi', showing kimi k2.6, kimi k2.7 code, and kimi k3, each with an angular stylized lowercase-k leading icon.

Unfiltered model picker scrolled to show the three kimi rows (angular-k icons) directly alongside minimax 2.7/3 and qwen 3.6/3.7 plus rows (bar-chart icons).

Model picker showing deepseek v4 pro and glm 5.2 (both bar-chart icons) alongside gemini rows (triangle-in-box icons) and gpt-5 rows (circle icons).

Model picker showing claude sonnet 5 rows (sparkle icons), deepseek v4 pro (bar-chart icon), and gemini rows (triangle-in-box icons).

Conversation: https://staging.warp.dev/conversation/d5d0a2ba-75ae-46fe-af56-5858937d1ce3
Run: https://oz.staging.warp.dev/runs/019fda75-5304-78cf-a0c0-87d624790025

This PR was generated with Oz.