feat(agent): add OrcaRouter as a named LLM provider preset

[putraperdana1207-pixel]

Summary

OpenChatCut's built-in agent is powered by the Vercel AI SDK, and its Agent LLM settings expose each model vendor as a named provider preset in shared/llm-providers.ts. Every preset is picked up automatically by the settings UI, the server-side keystore, model selection, and the /llm proxy — so adding a new OpenAI-compatible gateway is a small, well-defined wiring change.

This PR adds OrcaRouter as a first-class named provider preset, mirroring the existing OpenRouter wiring exactly: same openai-compatible protocol, same /chat/completions proxy path, same Bearer-key handling, same settings-page auto-generation.

OrcaRouter is an OpenAI-compatible AI gateway built for both models and agents. Like OpenRouter, it exposes a provider/model namespace across many models — but it also combines adaptive routing, automatic failover, zero-markup inference, observability, guardrails, and agent-tool governance behind the same endpoint. Adding orcarouter as a first-class provider means OpenChatCut users can use that stack directly, without treating OrcaRouter as an anonymous custom base URL.

What changed

File	Change
shared/llm-providers.ts	New orcarouter preset: protocol: 'openai-compatible', baseUrl: 'https://api.orcarouter.ai/v1', defaultModel: 'orcarouter/auto' — directly beside the openrouter preset it mirrors
server/keystore.ts	Whitelist LLM_ORCAROUTER_API_KEY, LLM_ORCAROUTER_BASE_URL, LLM_ORCAROUTER_MODEL (same three slots as every other provider)
src/components/settings/VisionModelPane.tsx	OrcaRouter label in the vision-provider picker
src/components/settings/vendorIcons.tsx	orcarouter vendor id + a monogram mark (no official SVG vendored yet, same treatment as xAI/BytePlus)
src/agent/ai-sdk.verify.ts	Coverage for the new preset: normalizeLlmProvider, defaultModelForProvider, providerApiPath, and getLanguageModel provider id (orcarouter.chat)
README.md / README_ZH.md	OrcaRouter added to the Agent row in the Architecture table

The runtime path needs no new code: createProviderFactory's default branch already constructs createOpenAICompatible({ name: provider }) for non-dedicated providers, so orcarouter resolves to the orcarouter.chat adapter exactly like openrouter.chat does, and server/plugins/llm-proxy.ts routes x-openchatcut-provider: orcarouter to the preset base URL with a Bearer key from the keystore.

How the provider is used

In Settings → Agent 模型 → Agent 大脑, OrcaRouter now appears as its own vendor page. Save an OrcaRouter API key, and the model picker offers orcarouter/auto (the adaptive default) or any vendor/model id the gateway exposes (e.g. anthropic/claude-opus-4.8, deepseek/deepseek-v4-pro).

Verification

• npx tsc -b --force — passes
• npm run lint — passes
• Targeted verifies for the touched paths (src/agent/ai-sdk.verify.ts, shared/model-capabilities.verify.ts, server/keystore.verify.ts, server/llm-proxy.verify.ts, src/agent/settings/agent-settings.verify.ts, server/key-probes.verify.ts) — all pass
• npm run build — passes
• Live check: a real round-trip through the /llm proxy helpers to https://api.orcarouter.ai/v1/chat/completions with model: orcarouter/auto returned HTTP 200 with a valid completion

The full npm test suite runs locally except the Remotion headless-browser segment, which needs a system libnspr4/libnss3 that this sandbox cannot install; the remaining suite (including every provider-path verify) is green, and CI's ubuntu runner installs those libraries.

It also runs gateway-level, zero-trust security for AI agents on the same endpoint — screening every prompt/response and governing every tool call on a default-deny basis, with no application code changes.

Discord: discord.gg/YEubt8enRA · X: https://x.com/OrcaRouter

I'm an engineer on the OrcaRouter team.