Captures d'écran du projet
Vue d’ensemble
NeuroLink is a TypeScript-first AI integration layer designed to route streams of tokens, tool calls, memory, and documents across various LLM providers and connectors. Extracted from production systems at Juspay (powering Tara, Yama, and Clairvoyance), it abstracts the complexity of interacting with multiple AI services—such as OpenAI, Anthropic, Google, and AWS Bedrock—into a single consistent interface. It supports streaming architectures, multi-provider failover, and intelligent routing, allowing developers to switch providers or models with a single parameter change.
Fonctionnalités clés
- Unified API for 30+ AI providers and 100+ models
- MCP-native integration (connect any MCP server via stdio or HTTP)
- Multi-provider voice support (6 TTS, 4 STT, 2 realtime APIs)
- Built-in RAG with 10 chunking strategies, hybrid search, and reranking
- Per-user condensed memory across conversations (S3, Redis, SQLite)
- Human-in-the-Loop (HITL) tool approval and output validation
- Provider fallback policies and model chaining
- Context window management with 4-stage compaction pipeline
- Structured output using Zod schemas
- File processor system supporting 17+ file types
- Professional CLI and TypeScript SDK
- AutoResearch autonomous experiment engine
Prérequis, installation et démarrage rapide
Utilisation
Compatibilité des modèles et cas d’usage
Supports 100+ models across 30+ providers including OpenAI, Anthropic, Google Vertex AI/AI Studio, AWS Bedrock, Azure, DeepSeek, NVIDIA NIM, Ollama, LM Studio, and llama.cpp.
Licence et notes sur les risques
Repository metadata declares MIT. Before publishing or commercial use, review the repository LICENSE file, copyright notices and dependency terms.
Editorial verification 2026-08-02: repository URL, owner, description, license and repository statistics were reviewed. License metadata: MIT. README was fetched for the channel draft; re-check repository dependencies, releases and model terms before production use.
Publication et maintenance
Q1 2026 updates include Avatar/Music modalities with 12 new providers, Multi-Provider Voice (TTS/STT) via generate/stream options, DeepSeek/NVIDIA NIM/LM Studio/llama.cpp providers, ModelAccessDeniedError, Provider Fallback Policy, Per-Request Credentials, AutoResearch, Gemini 3 Multi-turn Tool Fix, MCP Enhancements (tool routing, caching, batching), Memory, Context Window Management, Tool Execution Control, File Processor System, and RAG with generate()/stream().