Add Thinking Machines (Tinker) model provider support

[devin-ai-integration]

Summary

Adds Thinking Machines' Tinker as a selectable inference provider, wired through the same "bring your own key" path as Fireworks/OpenAI/Anthropic. Users add their key + model checkpoint under Settings ▸ Keys and pick Thinking Machines in the model picker (iOS list + Mac menu, both data-driven so no picker-specific code was needed).

The Tinker OpenAI-compatible API differs from the other hosted providers in one important way: there is no fixed model list — the "model" is a user-specific sampler checkpoint path (tinker://…/sampler_weights/…). So we store two values instead of one: THINKY_API_KEY and THINKY_MODEL_PATH, and the model's apiModelID resolves from the stored path at request time.

Endpoint (from the Tinker OpenAI-compatible docs):
https://tinker.thinkingmachines.dev/services/tinker-prod/oai/api/v1/chat/completions, Bearer auth, OpenAI-compatible wire format (so TinkerChat reuses OpenAIChat.wireMessages exactly like FireworksChat).

Changes

• KeyStore: new Key.thinky (THINKY_API_KEY) + Key.thinkyModelPath (THINKY_MODEL_PATH), new Service.thinky ("Thinking Machines") surfacing both keys in its editor; model path is treated as non-secret in maskedPreview.
• ModelSelection: new ModelProvider.thinky and ModelSelection.tinker.

  case .tinker:
      provider          → .thinky
      apiModelID        → KeyStore.shared.value(for: .thinkyModelPath)  // user-supplied
      contextWindowSize → nil        // varies per checkpoint
      requiredKey       → .thinky
      supportsVision    → false      // text turns only

• TinkerChat (new): streaming client mirroring FireworksChat; errors visibly when the key or model path is missing rather than silently falling back.
• AgentHarness: routes .thinky → TinkerChat.
• ModelPickerVC: footer copy for the Thinking Machines section.
• Exhaustive-switch updates so the new enum cases compile everywhere: ConversationTitleService (adds a Tinker title path, only used when Tinker is the selected provider and both values are set), VisionSummaryService (grouped with .apple → OCR fallback, since Tinker is text-only), VMAgentRuntime (.thinky falls through — the Loop VM runner can't target Tinker's custom base URL), and IntegrationSkill key-alias map.
• ContextUsageTests: .tinker joins .appleFoundation as a model with an intentionally-nil context window.
• Secrets.xcconfig.example: documents the two new build-time vars.

Onboarding was intentionally left out of scope — its simple single-key paste-chip flow doesn't fit Tinker's key + checkpoint-path requirement; Tinker is added via Settings ▸ Keys instead.

Testing

Not build-verified: this is a Swift/Xcode iOS+macOS app and the working environment has no macOS toolchain (swiftc/xcodebuild unavailable). Verified instead by auditing that every exhaustive switch over ModelProvider/ModelSelection now handles the new cases.

Link to Devin session: https://app.devin.ai/sessions/a2cdd7f461b1428ebbd7d589308818d5
Requested by: @theashbhat

[devin-ai-integration]

Original prompt from Ash Bhat

can you help me add support for thinking machine's new model

the secret for this will be THINKY_API_KEY
lets add support for this in settings and enable users to select this in model picker for iOS and Mac

here's the documentation
https://tinker-docs.thinkingmachines.ai/tinker/compatible-apis/openai/ @getathelas/LoopHarness

[devin-ai-integration]

🤖 Devin AI Engineer

I'll be helping with this pull request! Here's what you should know:

✅ I will automatically:

• Address comments on this PR. Add '(aside)' to your comment to have me ignore it.
• Look at CI failures and help fix them

Note: I can only respond to comments from users who have write access to this repository.

⚙️ Control Options:

• Disable automatic comment, CI, and merge conflict monitoring

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
loop-harness	Ready	Preview, Comment	Jul 16, 2026 3:59pm