Company Brain
An open-source enterprise knowledge assistant project that combines knowledge graphs, expert agents and large models to turn organizational knowledge into a conversational query assistant, supporting CSV data import and graph retrieval.
Timeline (last 90 days)
NexusOps uses two graphs: Company Brain (nexus_company_brain) stores the enterprise topology (microservices, caches, databases, git commits, runbooks, engineers), and Episodic Incident Memory (nexus_incident_memory) stores historical incidents, symptoms, and remediation playbooks
Storing the architecture topology graph and the incident memory graph separately avoids polluting the graph structure with historical events, while shared service identifiers preserve cross-graph query capability
Company Brain's backend uses Python + FastAPI to provide REST interfaces, and its frontend uses React + TypeScript to build the dashboard, including a Streamlit chat and CSV import interface.
Company Brain's question routing is currently keyword-based, dispatching questions to expert agents such as People & Teams, Projects, Issues & Telecom, Documents, and Decisions, and handing unmatched questions to the Generalist.
Company Brain calls LLMs through OpenRouter to generate answers.
Company Brain is an open-source project that combines a knowledge graph, expert agents, and large models to build an enterprise knowledge assistant.
Company Brain's currently optional Supabase chat history strategy allows anonymous read and write, and must be tightened and supplemented with authentication and role-based access control before real enterprise data goes live.
The Company Brain project documentation recommends introducing semantic routing in the future, evaluating answers with a test question set, keeping data fresh through scheduled imports, and introducing embedding-based semantic retrieval.
Company Brain's backend attaches the source and evidence information of graph records in its answers, making answers traceable.
Company Brain's data import supports employee information, engagement surveys, recruitment records, training data, and project data in CSV format.
All Facts (10)
Storing the architecture topology graph and the incident memory graph separately avoids polluting the graph structure with historical events, while shared service identifiers preserve cross-graph query capability
50%UnverifiedNexusOps uses two graphs: Company Brain (nexus_company_brain) stores the enterprise topology (microservices, caches, databases, git commits, runbooks, engineers), and Episodic Incident Memory (nexus_incident_memory) stores historical incidents, symptoms, and remediation playbooks
50%UnverifiedCompany Brain's backend uses Python + FastAPI to provide REST interfaces, and its frontend uses React + TypeScript to build the dashboard, including a Streamlit chat and CSV import interface.
50%UnverifiedCompany Brain's question routing is currently keyword-based, dispatching questions to expert agents such as People & Teams, Projects, Issues & Telecom, Documents, and Decisions, and handing unmatched questions to the Generalist.
50%UnverifiedCompany Brain calls LLMs through OpenRouter to generate answers.
50%UnverifiedCompany Brain is an open-source project that combines a knowledge graph, expert agents, and large models to build an enterprise knowledge assistant.
50%UnverifiedCompany Brain's currently optional Supabase chat history strategy allows anonymous read and write, and must be tightened and supplemented with authentication and role-based access control before real enterprise data goes live.
50%UnverifiedCompany Brain's backend attaches the source and evidence information of graph records in its answers, making answers traceable.
50%UnverifiedCompany Brain's data import supports employee information, engagement surveys, recruitment records, training data, and project data in CSV format.
50%UnverifiedThe Company Brain project documentation recommends introducing semantic routing in the future, evaluating answers with a test question set, keeping data fresh through scheduled imports, and introducing embedding-based semantic retrieval.
50%