feat: add ModelRunner as a chat provider

[hakankaan]

Upfront disclosure: I work on ModelRunner, so this PR adds my own company as a provider. Close it without ceremony if that isn't something you take from outside.

On your AI usage policy — this was written with Claude Code, and I've read every line of it, traced the code paths it touches, and run your bun run build against it. It's ~75 added lines across 11 files with no new abstraction, so it's small enough to maintain and small enough to review honestly. If any part of it reads as generated-and-not-understood, say so and I'll explain the reasoning or drop it.

What this adds

ModelRunner speaks the OpenAI chat format, so it needs no new adapter. OpenAIAdapter already composes {base}/chat/completions from a base URL ending in /v1, and the trait's default list_models_endpoint / parse_models_list handle the discovery list unchanged.

Provider id	modelrunner
Base URL	https://queue.modelrunner.run/v1
Adapter	openai::OpenAIAdapter
Auth	Authorization: Bearer <key>
Key page	https://modelrunner.ai/settings/api-keys

Everything else is the wiring one provider needs — the config tuple, the onboarding default base URL in both reducers, the aggregator description bucket, the icon, the parameter-support profile, and the key-location guide. All 11 files are additive; no existing line changes behaviour.

Choices worth flagging

Three of these are deliberate omissions, which are easier to mistake for oversights than the additions are:

• Not added to supports_model_list in models/verify.rs. That list makes the save path check the model against GET /models and refuse on a miss. ModelRunner's discovery list is public models only, while a key can also call private ones — so adding it there would block a user from saving a model they legitimately own, with "Model not found on provider". The default (skip the check, assume it exists) is correct for this provider, not an oversight.
• The dispatch arm is written out rather than left to _ => OpenAIAdapter. The fallback would already do the right thing, but your Backend Guidelines ask for explicit behaviour over hidden fallback logic, and lettuce-host sets the precedent of naming it anyway.
• reasoningSupport: "none". The profile mirrors your Mistral entry: the surface forwards the standard OpenAI sampling parameters (temperature, top_p, max_tokens, frequency_penalty, presence_penalty) to the model and exposes no reasoning-effort toggle of its own. Claiming "effort" or "dynamic" would put a control in the UI the backend can't honour.
• Guide copy is English only. translateWithLocale() falls back to the English tree for a missing key, so all 21 locales render it correctly today. The other 20 translations should come from speakers of those languages, not from me — I'd rather leave a gap your process can fill than seed 20 machine translations into the locale files.
• Grouped with the aggregator description (nanogpt / featherless / openrouter) in both ProviderSetup.tsx and ProviderCard.tsx, since it's one key over many models rather than a single lab's own API.
• Placed after anannas in config.rs, among the hosted providers rather than at the top of the list.

One thing I noticed and did not touch, since your CONTRIBUTING asks to avoid unrelated refactors: the default base URL for a provider is now spelled in three places — config.rs, providerReducer.ts and onboardingReducer.ts — and the two reducers are byte-identical switches. Happy to open a separate PR folding them into one if you'd find that useful.

Testing

• bun install --frozen-lockfile && bun run build (i.e. tsc && vite build) — exit 0, which is your PR Check "Frontend Check" job. tsc --noEmit covers all nine TypeScript edits, including the TranslationKey union derived from en.ts.
• The icon import resolves and bundles: dist/assets/modelrunner-YrAT_Kko.png.
• GET https://queue.modelrunner.run/v1/models answers 401 where every neighbouring GET path answers 404 (/v1/nonsense, /v1/models/extra) — the route exists and authenticates, which is what build_verify_url's {base}/models default branch and list_models_endpoint both target.
• The Rust side is two lines: a (&str, &str, &str) tuple in the existing vec! and a match arm identical in shape to the lettuce-host one directly below it. I don't have a Rust toolchain on this machine, so cargo check was not run locally — your Linux/macOS/Windows/Android check jobs will cover it.

---

🤝 Partnership & contact

This PR comes from the ModelRunner team. Two things beyond the code, in case they're useful:

• Free API credits for this project. We run an Open Source Program that grants credits to maintainers of open-source projects — the tiers are on the page, and this project sits comfortably inside them. That offer stands whether or not you merge this PR; it isn't conditional on the integration.
• Happy to revise anything here to match your conventions — naming, structure, scope, or splitting it up. Leave a comment and I'll push a change.

Reach us anytime:

• 📧 support@modelrunner.ai
• 🌐 https://modelrunner.ai
• 📚 Docs: https://modelrunner.ai/docs

🤖 Generated with Claude Code

[Husky110]

Hmm - Looking that this, two things catch my eye:
1.) Please also provide the language strings for the other languages.
2.) I like your offering about free credits, however: Looking at you website I was unable to locate where your company is located. There are e-mailaddresses, but if shit hits the fan, where should I file my complaint about a databreach or a privacy-concern?
I would kindly ask you for this information. :)

[hakankaan]

Both fair.

1. Languages — you're right, English-only made it the odd one out. I'll push the remaining 20 here, keeping "ModelRunner" untranslated.

Two pre-existing ones I hit while matching the register — happy to fix in this PR or leave for a separate one: tr.ts translates the brand, so s1 reads "tüysüz.ai'yi ziyaret edin" and points at a site that doesn't exist; and ru.ts s2 is truncated mid-sentence.

2. Where we are — ModelRunner, Inc., 1111B S Governors Ave # 95898, Dover, DE 19904, United States. The Terms put governing law in California, venue San Francisco.

You were right that it wasn't findable, and that's now fixed: the address is on https://modelrunner.ai/privacy and https://modelrunner.ai/terms. Privacy contact is support@modelrunner.ai.

[MegalithOfficial]

Thanks a lot for taking the time to put this together, and especially for being upfront about your connection to ModelRunner.

For now, we're going to pass on adding ModelRunner as a built-in provider. We're trying to keep the officially supported provider list fairly limited, and we generally only add new providers when there's a clear need from our user base or a strong reason for us to maintain a dedicated integration.

This isn't related to the quality of the PR. You clearly put a lot of effort into keeping the implementation small and consistent with the existing provider setup, and we appreciate that.

Since ModelRunner is OpenAI-compatible, users who want to use it can still add it through LettuceAI's custom provider support without needing a built-in integration.

Thanks again for the contribution and for the thoughtful PR.