feat: Convert LangChain implementation to new AIProvider interface

[jsonbailey]

Note

Refactors LangChain integration to an AIProvider-based provider with a new invoke flow and metrics, removes TrackedChat, and updates tests and configs.

• Core (LangChainProvider)
  • Convert to AIProvider subclass with constructor-held BaseChatModel, invokeModel, and getChatModel.
  • Add static create(aiConfig) and createLangChainModel factory using initChatModel and mapProvider.
  • Replace createTokenUsage/tracking helpers with createAIMetrics returning { success, usage }.
  • Keep convertMessagesToLangChain and provider mapping (e.g., gemini → google-genai).
• Removal
  • Delete LangChainTrackedChat and its exports; index.ts now exports only LangChainProvider.
• Tests
  • Update message imports to @langchain/core/messages; mock langchain/chat_models/universal.
  • Add tests for createAIMetrics and mapProvider; adjust expectations and formatting.
• Config
  • Update Jest config: use transform, broaden testMatch, simplify coverage settings, and add moduleFileExtensions.
  • Adjust ESLint tsconfig includes/excludes.

^{Written by Cursor Bugbot for commit fe29f8f. This will update automatically on new commits. Configure here.}

[cursor]

Bug: ESLint Include Pattern Uses Absolute Path

The include pattern ["/**/*.ts"] in tsconfig.eslint.json uses a leading slash, which makes it an absolute path from the filesystem root. This prevents ESLint from finding TypeScript files within the package, as it expects a relative path from the project directory.

 

[knfreemLD]

Just a thought: should this be promptTokens + completionTokens instead? I don't know if totalTokens would ever be different than prompt+completion, but if it ever was I could see users being confused

[jsonbailey]

I am trying to find the conversation where this came up before. I believe @kinyoklion mentioned it. We used total tokens because it is possible some models may use additional tokens not accounted for in just the prompt / completion tokens.

[kinyoklion]

Yeah, that is the thought. We originally didn't know that prompt + completion = total would be a guarantee. Which seems more likely if we eventually care about thinking tokens.

For example total may become prompt + thinking + completion in which case our total would be incorrect if it was just prompt + token.

[kinyoklion]

Though some providers currently aren't charging for thinking so they are excluding it from their total calculation. But I could see it change either way.