Cognizant’s multi-agent studio: HOCON in, agent networks out

What it does

Neuro SAN Studio is a companion playground for Cognizant’s Neuro SAN orchestration framework, bundling demo agent networks, tutorials, and a web UI called nsflow. You declare agent behavior and topology in HOCON files, and the framework coordinates how LLM-powered agents split up and delegate work. It is largely a demo and tooling layer on top of the core library, aimed at getting an agent network from description to running server quickly.

The interesting bit

Rather than hard-coding a control flow, the framework uses an AAOSA protocol that lets agents autonomously decide who handles which subtask. A meta-agent called the Agent Network Designer can read a plain-English use-case description and emit a new HOCON config for you. There is also a Sly-Data mechanism that keeps sensitive fields out of LLM prompts during handoffs between agents.

Key highlights

• Entire agent networks defined declaratively via HOCON instead of orchestration code.
• Meta-agent generates new multi-agent configs from natural language descriptions.
• Sly-Data hides sensitive information from models during inter-agent transfers.
• Integrates with LangChain tools, MCP, A2A, CrewAI, Agentforce, and external APIs.
• Example networks cover banking, insurance, airline support, retail, and telco scenarios.

Verdict

Worth exploring if you want an opinionated, configuration-first multi-agent stack with enterprise industry templates. Give it a pass if you are looking for a lightweight, code-first agent library without the scaffolding and UI overhead.