neurolink
One TypeScript interface for 24+ LLM providers — swap providers without rewriting. MCP-native (connect any MCP server), voice (TTS/STT/realtime), RAG, memory, file processors. Production-origin: powers Tara, Yama, and Clairvoyance at Juspay.
NeuroLink unifies 30-plus LLM providers behind one TypeScript interface
juspay/neurolink is a TypeScript library and CLI that wraps more than 30 AI providers and 100 models in a single streaming API. It adds MCP tooling, voice, RAG, memory, and human-in-the-loop controls.
One interface for many providers
NeuroLink presents itself as a universal AI integration platform that unifies 30 or more providers and 100 plus models under one consistent API. The README lists OpenAI, Anthropic, Google, AWS Bedrock, and Azure among the supported providers, and notes that switching providers is a single parameter change. Everything is treated as a stream, so text generation and token streaming share the same code path regardless of the backend chosen.
Beyond text generation
The library covers more than chat. It connects to any MCP-compliant tool server, supports multi-provider voice with TTS and STT providers plus realtime APIs, and performs RAG by chunking, embedding, and searching documents. Memory is kept per user across conversations with LLM-powered condensation backed by S3, Redis, or SQLite. A file processor system handles 17 or more file types, and a provider fallback policy walks a model chain when access is denied.
Built for production use
The README says NeuroLink was extracted from production systems at Juspay, where it powers products named Tara, Yama, and Clairvoyance. For regulated work it includes a human-in-the-loop system with tool approval workflows, output validation, confidence thresholds, and an audit trail aimed at HIPAA, SOC 2, and GDPR programs. An AutoResearch feature runs autonomous experiment loops that propose code changes, run experiments, and evaluate metrics unattended.
Editorial conclusion
The package is published on npm as @juspay/neurolink and is released under the MIT license.
Community notes