How cognizant-ai-lab/neuro-san works
Neuro SAN is a Python library and service for defining and running data-driven multi-agent networks. The primary workflow is a user sending a prompt through a direct or HTTP client; the selected network’s front-man coordinates agent/tool work and returns a response. Network configuration, LLM-provider policies, coded tools, messaging, MCP, authorization, and service-side network updates are distinct capabilities. The graph combines README-documented behavior with relationships visible in the supplied source index and excerpts; unsampled implementation links are omitted where evidence is unavailable.
Main entry points: agent_session_factory.py Session factories · async_agent_service.py Agent service · agent_cli.py Agent CLI · http_service_agent_session.py HTTP session
Components 24
Client and sessions
- Agent CLI
neuro_san/client/agent_cli.py - Session factories
neuro_san/client/agent_session_factory.py - Direct session
neuro_san/session/direct_agent_session.py - HTTP session
neuro_san/session/http_service_agent_session.py - Session contract
neuro_san/interfaces/agent_session.py
Network runtime
- Agent network
neuro_san/internals/graph/registry/agent_network.py - Config resolution
neuro_san/internals/graph/filters/resolved_network_config_filter.py - Agent activations
neuro_san/internals/graph/activations - Chat session
neuro_san/internals/chat/data_driven_chat_session.py - Message processing
neuro_san/message/processors
Models and tools
- LLM providers
neuro_san/internals/run_context/langchain/llms - Tool runtime
neuro_san/internals/run_context/langchain/tools - Coded tools
neuro_san/coded_tools
Service and network management
- MCP service
neuro_san/service/mcp - HTTP service
neuro_san/service/http/server/http_server.py - Agent service
neuro_san/service/generic/async_agent_service.py - Authorization
neuro_san/service/authorization - Network storage
neuro_san/internals/network_providers/agent_network_storage.py - Manifest restore
neuro_san/internals/graph/persistence/registry_manifest_restorer.py - Registry updater
neuro_san/service/watcher/registries/registry_storage_updater.py - Reservations
neuro_san/internals/interfaces/reservations_storage.py - Temporary deployments
neuro_san/service/watcher/temp_networks/updater/temp_network_storage_updater.py - Reservation data
neuro_san/interfaces/reservation.py
Other components
- User
Connections 22
- User → Agent CLI: submits prompt
README.md - Agent CLI → Session factories: creates session
- Session factories → Direct session: opens direct
- Session factories → HTTP session: opens HTTP
- Direct session → Session contract: implements
- HTTP session → Session contract: implements
neuro_san/session/http_service_agent_session.py - Agent CLI → HTTP session: uses HTTP
README.md - Session factories → Agent network: loads network
- Agent network → Config resolution: resolves config
- Chat session → Agent network: runs network
- Chat session → Message processing: processes messages
- Agent activations → LLM providers: calls models
- Agent activations → Tool runtime: dispatches tools
- Tool runtime → Coded tools: creates tools
README.md - HTTP service → Agent service: serves agents
- Agent service → Network storage: gets networks
- Agent service → Authorization: checks access
- Agent service → MCP service: exposes MCP
README.md - Registry updater → Manifest restore: restores manifest
neuro_san/service/watcher/registries/registry_storage_updater.py - Registry updater → Network storage: updates networks
neuro_san/service/watcher/registries/registry_storage_updater.py - Temporary deployments → Reservations: deploys reservations
neuro_san/service/watcher/temp_networks/updater/temp_network_storage_updater.py - Reservations → Reservation data: stores records
neuro_san/internals/interfaces/reservations_storage.py