Network-AI

What Network-AI does

The core is a shared blackboard with locking. Writes go through an atomic propose, validate, commit sequence, which the README says prevents last-write-wins overwrites and split-brain state across parallel agents. Governance covers FSM rules, per-agent token ceilings, HMAC and Ed25519 audit trails and permission gating. A ContextComposer assembles token-budgeted, relevance-ranked context packs, and the context_pack and blackboard_search MCP tools let agents pull curated state instead of the whole board.

Network-AI can be used as a library (createSwarmOrchestrator), an MCP server (npx network-ai-server --port 3001), a CLI, a Claude Code plugin, a Gemini CLI extension or an OpenClaw skill. Its 32 adapters include LangChain, AutoGen, CrewAI, OpenAI Assistants, LlamaIndex, MCP, OpenClaw, Codex and a Hermes adapter that targets NousResearch Hermes or any OpenAI-compatible endpoint. A GovernedModelGateway handles model fallback, repricing and auditing behind one budgeted call.

Key features

• Shared blackboard with atomic propose, validate, commit locking
• Per-agent token ceilings, FSM governance and signed audit trails
• ContextComposer with context_pack and blackboard_search MCP tools
• 32 adapters, including LangChain, CrewAI, AutoGen, OpenClaw and Hermes
• Usable as a library, MCP server, CLI, Claude Code plugin or Gemini CLI extension
• GovernedModelGateway for fallback and budgeted, audited model calls

When to use it

• Stopping parallel agents from overwriting each other's results in shared state
• Putting token budgets and permission checks around a mixed-framework agent system
• Coordinating agents from different frameworks through one orchestrator

Who it is for: TypeScript and Node.js developers who run several agents, possibly from different frameworks, and need coordination and governance.

How it fits with Hermes Agent

Supports Hermes among its 32 adapters, through an adapter for NousResearch Hermes or any OpenAI-compatible endpoint, and the repository is tagged hermes-agent.

FAQ