[Suggestion] LangChain.js audio transcription tool and document loader (TypeScript)

[deepgram-robot]

What to build

A LangChain.js integration that provides Deepgram as a tool and document loader for TypeScript/JavaScript AI agent applications — enabling LangChain agents to transcribe audio files, process audio URLs, and load audio content as documents for RAG pipelines.

Why this matters

LangChain.js is one of the most widely-used AI agent frameworks for TypeScript/JavaScript developers. Developers building agentic applications need audio transcription as a tool their agents can invoke — for example, an agent that processes voice memos, analyzes podcast episodes, or handles audio uploads. A Python LangChain integration exists, but there is no equivalent for LangChain.js, leaving JavaScript/TypeScript developers without an official path to add Deepgram audio capabilities to their LangChain agents.

Suggested scope

• Language: TypeScript
• Framework: LangChain.js (@langchain/core)
• Deepgram APIs: Pre-recorded STT, optionally Streaming STT
• What to include:
  • DeepgramTranscriptionTool — LangChain StructuredTool for pre-recorded audio transcription (file path or URL input)
  • DeepgramAudioLoader — LangChain DocumentLoader for loading audio files as transcribed documents
  • Example agent that uses the transcription tool to process and summarize audio
  • Support for key STT parameters (model, language, diarization, smart_format)
  • Full TypeScript types

Acceptance criteria

• Runnable with minimal setup (clone, add API key, npm install, run)
• README explains the integration pattern and LangChain.js compatibility
• Uses current Deepgram JS SDK and @langchain/core versions
• Tool is compatible with LangChain.js agent executor (createToolCallingAgent)
• Document loader produces properly structured LangChain Document objects
• Includes a working example agent that transcribes and summarizes audio

---

Raised by the DX intelligence system.