Neuroverse

What is Neuroverse?

Neuroverse is an MCP server that gives AI agents multilingual intelligence (Tamil, Hindi, Telugu, Kannada, Malayalam, Bengali + English code-switching), tiered memory, a voice layer, intent extraction, multi-model routing, and three-layer safety. It is designed for autonomous agents that need to understand mixed Indian languages, remember context across sessions, and execute actions safely.

How to use Neuroverse?

Install via npm (npm install neuroverse) for a Node.js 18+ environment (zero database deps) or from source (Python 3.10+, requires PostgreSQL for persistent memory). Add the standard MCP stdio configuration to your host (Cursor, VS Code Copilot, Claude Desktop). Then instruct your agent to use MCP tools like neuroverse_process for handling requests and neuroverse_store/neuroverse_recall for persistent context.

Key features of Neuroverse

• Multilingual Intelligence (Vani) with keyword normalisation for code-switched input
• Intent Extraction (Bodhi) — LLM-first with deterministic rule-based fallback
• Tiered Memory (Smriti) — short-term, episodic, semantic with importance scoring
• 3-Layer Safety (Kavach) — blocklist → risk classifier → LLM judge at zero token cost
• Multi-Model Router (Marga) — routes tasks to OpenAI, Anthropic, Sarvam AI, Ollama, OpenRouter
• Agent-to-Agent (Setu) — REST+JSON agent registry for automatic handoff

Use cases of Neuroverse

• Handling user queries in mixed Indian languages (e.g., "anna indha file ah csv convert pannu")
• Maintaining agent context across chat sessions via tiered memory
• Blocking dangerous actions like DROP DATABASE before execution
• Routing analytical reasoning tasks to OpenRouter and cheaper tasks to local models
• Enabling voice input (Whisper STT) and voice output (Coqui TTS) for agents

FAQ from Neuroverse

What languages does Neuroverse support?

Neuroverse supports Tamil, Hindi, Telugu, Kannada, Malayalam, and Bengali with English code-switching. It normalises only domain-critical keywords rather than full translation.

Does Neuroverse require a database?

The npm edition uses JSON files and requires no database. The Python source edition requires PostgreSQL for persistent tiered memory.

How does the safety layer work?

Kavach uses three layers: a keyword blocklist, an intent risk classifier, and an LLM judge. It runs at zero token cost and sub-millisecond latency, blocking dangerous actions before execution.

What model providers can Neuroverse route to?

The multi-model router (Marga) supports OpenAI, Anthropic, Sarvam AI, Ollama, and OpenRouter. It selects the best model automatically based on task type.

What are the system requirements?

Node.js 18+ for the npm edition; Python 3.10+ with PostgreSQL for the source edition. No other runtime dependencies are mandatory.