add OpenRouter provider preset

[kenrogers]

Why

OpenRouter already works through Open Design's OpenAI-compatible proxy when users know the custom base URL, but it was not available as a quick-fill provider in Settings. Adding a preset makes the existing BYOK path discoverable without introducing a new adapter or dependency.

What changed

• Added OpenRouter to the known OpenAI-compatible provider presets with https://openrouter.ai/api/v1 and current model suggestions.
• Documented the OpenRouter API-mode setup in QUICKSTART.md.
• Added focused coverage for the OpenRouter preset and OpenAI-compatible routing.

Validation

• pnpm install
• pnpm --filter @open-design/web test -- apps/web/tests/components/SettingsDialog.test.ts apps/web/tests/providers/openai-compatible.test.ts
• pnpm --filter @open-design/web typecheck
• pnpm guard
• pnpm --filter @open-design/desktop build
• pnpm typecheck

[lefarcen]

Hi @kenrogers! 🎉
Thanks for the contribution — adding OpenRouter as a preset makes BYOK more discoverable.
I will run a deep review and get back to you within 24h.

Thanks for making open-design better!
— open-design team

[lefarcen]

Hey @kenrogers! Thanks for adding OpenRouter — this makes BYOK more discoverable. 👍

I've reviewed the config, tests, and docs. The core implementation is solid, but I found a few gaps in test coverage and documentation that would help users avoid setup pitfalls:

Main findings:

• The settings test doesn't actually prove the OpenRouter preset appears in the UI dropdown
• Model IDs may not match the token-limit lookup, which could truncate large artifacts
• Docs don't mention where to get an API key or warn about OpenRouter-specific account/rate-limit errors

See inline comments for details. None of these block merge, but addressing them would strengthen the feature.

[lefarcen]

P2: This test manually patches baseUrl/model, so it would still pass if OpenRouter were removed from KNOWN_PROVIDERS. Consider asserting the KNOWN_PROVIDERS entry or simulating actual provider selection from the Settings UI to prove discoverability.

[lefarcen]

P2: The model IDs here (openai/gpt-5.2, openrouter/auto, etc.) may not match modelMaxTokensDefault lookup keys in litellm-models.json. If they fall back to 8192, large artifact responses could be truncated. Consider explicit overrides or OpenRouter-aware normalization.

[lefarcen]

P3: This routing test only covers the isOpenAICompatible heuristic. It doesn't prove the OpenRouter preset actually flows through /api/proxy/openai/stream at runtime. An integration-style provider test using the preset config would catch regressions.

[lefarcen]

P3: Docs don't tell users where to get an OpenRouter API key or that account credits/free-model availability can affect success. New users hitting 401/402/429 won't have enough context to distinguish bad app config from OpenRouter account/rate-limit issues.

[lefarcen]

P3: Missing manual-setup warning: if a user pastes https://openrouter.ai/api/v1 while still on Anthropic/custom protocol, requests could go through the wrong proxy shape and fail. Add a note that protocol must be OpenAI.

[PerishCode]

@kenrogers — checking in on the OpenRouter preset PR. It's been about 4 days since lefarcen left feedback and we're holding for your next iteration. Still in flight? Drop a comment if something in the review thread needs clarification, or let us know if you'd like to close this in favor of another approach.

(Tagged awaiting-author-response-24h by maintainer tooling.)

[kenrogers]

Thanks for the review. I pushed c20f6ce with updates for the feedback:

• assert the OpenRouter entry in KNOWN_PROVIDERS so the preset is covered directly
• normalize OpenRouter model IDs through the vendored LiteLLM openrouter/... aliases so preset models avoid the 8192 fallback
• add an integration-style provider test proving the OpenRouter preset posts to /api/proxy/openai/stream with the expected token default
• expand the Quickstart with the OpenRouter keys link, account/credit/rate-limit note, and the manual setup warning to stay on the OpenAI protocol tab

Validation passed locally: web tests, web typecheck, pnpm guard, and root pnpm typecheck.