
CompanyOS: Building an AI-Powered Company Brain with Knowledge Graphs
CompanyOS is an AI-powered company knowledge assistant that connects employees, teams, projects, tasks, meetings, decisions, and documents into a unified knowledge graph. Built using React, FastAPI, FalkorDB, and Google Gemini, it enables users to ask questions in natural language, explore organizational relationships, and retrieve context-aware answers supported by relevant source documents. CompanyOS aims to simplify knowledge discovery, improve project visibility, and help teams make informed decisions.
Introduction
Every company generates knowledge every day through meetings, project updates, tasks, decisions, internal documents, and policies. However, this information is often scattered across different tools and files, making it difficult to find the right information at the right time.
Imagine joining a team and asking:
- Why was a project deadline changed?
- Which tasks are blocking the current release?
- What decisions were made during the last meeting?
- Who is responsible for a particular task?
- Which document explains a specific business decision?
Answering these questions may require searching through multiple records and connecting information manually.
That led me to explore an interesting question:
What if a company could have its own AI brain that understands how its people, projects, tasks, meetings, and decisions are connected?
That idea became CompanyOS — an AI-powered Company Brain.
CompanyOS explores how artificial intelligence, graph databases, and natural-language interfaces can work together to make organizational knowledge more accessible, connected, and useful.
The Problem: Company Knowledge Is Connected, but Our Tools Often Aren't
Traditional search systems are useful when we know exactly what we are looking for. But answering complex business questions often requires information from multiple sources.
Consider a project called Phoenix. One of its tasks is delayed because it depends on another incomplete task. A meeting note explains the dependency, while a decision record describes why the deadline was changed.
These records are related, but they may be stored separately.
A keyword search might find individual documents without explaining the complete situation.
The real challenge is not simply finding information. It is understanding the relationships between different pieces of information.
CompanyOS addresses this problem by representing company knowledge as a connected graph and providing an AI assistant that can retrieve relevant context and explain it in natural language.
The Idea Behind CompanyOS
CompanyOS is designed as a conversational knowledge assistant for organizations.
Instead of asking employees to remember where information is stored, the system aims to let them ask questions in everyday language and retrieve relevant company context.
For the prototype, I use a fictional software company called Nexora Technologies. Its organizational knowledge includes employees, teams, projects, tasks, meetings, decisions, policies, and source documents.
Using fictional company data allows the system to demonstrate realistic organizational scenarios without exposing confidential business information.
The central idea is simple: connect organizational knowledge so users can explore not only individual records, but also the relationships between them.
Key Features
1. AI Assistant — Ask the Graph
The AI assistant lets users ask questions about organizational knowledge using natural language.
For example:
- Which projects are currently delayed?
- Who owns Project Aurora?
- Which tasks are blocked or in progress?
- What decision explains a project's requirements?
- Which source document supports the answer?
The system retrieves relevant company records and uses the available context to generate an understandable response.
2. Knowledge Explorer
The Knowledge Explorer provides a visual representation of relationships between company entities.
Users can explore how employees, teams, projects, tasks, meetings, decisions, and documents connect to one another.
This helps make complex relationships easier to inspect than viewing isolated records in separate tables.
3. Project Intelligence
The project views bring together information about project status, ownership, progress, priorities, and task dependencies.
The aim is to make it easier to understand what is happening in a project and which related records may require attention.
4. People and Teams
Organizational relationships can be represented through team membership, task ownership, and project responsibilities.
This creates a foundation for questions about who is responsible for particular work and how teams contribute to projects.
5. Meetings, Decisions, and Documents
Meeting notes, decision records, and source documents provide context behind organizational activities.
Connecting these records to projects and tasks can help users investigate why a decision was made or which document supports a particular conclusion.
Technology Stack
CompanyOS combines modern web development, backend APIs, graph-based data storage, and AI.
Frontend: React and Vite
React provides the interactive user interface, while Vite supports the frontend development workflow.
Backend: Python and FastAPI
FastAPI provides the backend API layer and coordinates application operations, graph retrieval, and communication with the AI model.
Graph Database: FalkorDB
FalkorDB stores organizational entities and their relationships. Its graph-based model allows the application to represent connections between people, projects, tasks, meetings, decisions, and documents.
AI Model: Google Gemini
Gemini is used to generate natural-language responses based on the context retrieved by the application.
Local Development: Docker
Docker provides a convenient way to run FalkorDB in a local development environment.
Together, these technologies provide the foundation for an AI assistant that combines structured organizational knowledge with natural-language interaction.
How CompanyOS Works
The application follows a knowledge retrieval and answer-generation workflow.
Step 1: Organize Company Information
Company information is represented through structured records for employees, teams, projects, tasks, meetings, decisions, and documents.
Each record contains information that can be connected to other records.
Step 2: Build the Knowledge Graph
The system represents entities as nodes and their relationships as edges.
For example:
Employee → Assigned to → Task → Belongs to → Project
A project can also be connected to meeting notes, decisions, and source documents.
These relationships allow the application to represent organizational knowledge as a connected structure.
Step 3: Ask a Question
A user submits a natural-language question through the AI assistant.
For example:
“How is Project Aurora progressing, who owns it, which teams support it, and what tasks are blocked?”
Step 4: Retrieve Connected Information
The backend identifies relevant entities and relationships in the graph database.
It retrieves the available records needed to answer the question, such as project details, ownership information, task status, and related source documents.
Step 5: Generate an Answer
The retrieved context is provided to the AI model, which generates a readable response.
The application aims to ground its answers in the available organizational records rather than relying exclusively on the model's general knowledge.
Step 6: Present Supporting Sources
Where available, the assistant displays supporting source documents and related records so users can inspect the information behind the response.
This is important because an internal knowledge assistant should help users understand not only the answer, but also the evidence supporting it.
System Architecture
The application separates the frontend, backend, graph database, and AI generation components.
High-level workflow:
User Interface → FastAPI Backend → Graph Retrieval → FalkorDB → Retrieved Context → Google Gemini → Answer with Sources
The frontend sends the user's question to the backend. The backend retrieves relevant company information from FalkorDB and prepares the context required for the AI model. Gemini then generates a natural-language response, which is returned to the frontend.
This modular architecture allows individual components to evolve independently.
For example, graph queries can be optimized without completely redesigning the user interface. Additional data sources and retrieval methods can also be introduced as the project grows.

Project Screenshots
The following screenshots showcase the CompanyOS interface, graph visualization, and AI-powered question-answering experience.
1. CompanyOS Dashboard

The dashboard presents an overview of company activity, project progress, open tasks, graph information, and recent decisions.
It provides a starting point for understanding the current state of the organization's knowledge and work.
2. AI Assistant — Ask the Graph

The AI assistant answers natural-language questions about company information.
In the demonstrated example, the assistant connects project progress, ownership, supporting teams, task status, and source documents to provide a contextual answer.
3. Knowledge Explorer

The Knowledge Explorer visualizes organizational relationships as an interactive network.
It helps users investigate how people, teams, projects, tasks, decisions, and documents connect.
4. FalkorDB Graph Visualization

The FalkorDB interface shows the underlying graph structure, including node labels, relationship types, and connections between organizational entities.
This provides a view of the graph data that supports CompanyOS's knowledge retrieval capabilities.
A Practical Example: Investigating a Project
Imagine a team member asks:
“Why is Project Aurora delayed, who owns it, and which tasks are blocking progress?”
Answering this question may require several connected records:
- The project's status, priority, and deadline.
- The employee responsible for the project.
- Teams supporting the project.
- Tasks that are blocked or still in progress.
- Related decisions and source documents.
CompanyOS can retrieve the relevant connected records and use them as context for an AI-generated answer.
The user can then inspect the supporting source documents to understand where the information came from.
This example demonstrates the core idea behind CompanyOS: organizational questions often require connected knowledge, not just a matching document.
Technical Challenges and Design Considerations
Building a useful AI knowledge assistant involves more than connecting a language model to a database.
1. Representing Relationships Correctly
Organizational data contains different types of relationships. A task may belong to a project, depend on another task, or be assigned to an employee.
Representing these relationships consistently is essential because incorrect connections can produce misleading results.
A graph database provides an explicit way to model and query these relationships.
2. Retrieving Relevant Context
A knowledge graph can contain many records unrelated to a user's question.
Retrieving too little information can result in incomplete answers, while retrieving too much can introduce irrelevant context.
The retrieval process therefore needs to identify relevant entities and relationships before passing information to the AI model.
3. Generating Trustworthy Answers
Language models can generate plausible statements even when the required information is missing.
For an internal company assistant, unsupported answers can be particularly problematic.
The system should prioritize retrieved evidence, make supporting sources accessible, and acknowledge when the available information does not establish an answer.
4. Separating Retrieval from Generation
The graph database and the language model have different responsibilities.
FalkorDB represents and retrieves connected organizational knowledge, while Gemini generates natural-language responses using the retrieved context.
Separating these responsibilities makes the architecture easier to understand, test, and improve.
5. Handling Real-World Organizational Data
A production system would also need to handle changing records, duplicate information, access permissions, and sensitive documents.
These considerations are important for ensuring that the assistant remains reliable as the amount of organizational information grows.
Why Knowledge Graphs Matter for AI
Knowledge graphs provide a structured representation of entities and relationships.
This is useful when a question requires connecting multiple pieces of information rather than retrieving one isolated record.
For example, an employee may own a task, the task may belong to a project, and a meeting may explain a decision that affects the task.
A graph-based representation allows these connections to be queried explicitly.
When combined with an AI model, the graph can provide structured context that helps the assistant produce more relevant answers.
However, the quality of the result still depends on data quality, retrieval accuracy, and the model's ability to use the supplied context correctly.
Lessons Learned
Building CompanyOS reinforces several important principles for developing AI applications.
Good data modeling matters. The usefulness of an assistant depends on the quality and consistency of the information it can access.
AI is only one part of the system. Retrieval, data storage, application logic, and validation are equally important.
Relationships provide context. A knowledge graph can help represent connections that are difficult to understand when records are viewed independently.
Trust should be a design priority. Answers should be grounded in available information, with supporting evidence provided wherever possible.
Modular architecture supports experimentation. Separating the frontend, backend, graph database, and AI layer makes it easier to improve individual components.
Future Scope
There are several opportunities to extend CompanyOS.
- Integrate with real organizational knowledge sources.
- Introduce hybrid graph and vector retrieval.
- Improve source citations and evidence inspection.
- Add authentication and role-based access control.
- Support incremental updates when company information changes.
- Evaluate answer accuracy and retrieval quality using a dedicated test dataset.
- Explore project risk analysis and decision-impact discovery.
These improvements could help move the prototype toward a more comprehensive enterprise knowledge platform.
Security, access control, and data privacy would be especially important before connecting real company information.
Project Links and Demo
Explore the implementation and see the project in action.
Conclusion
CompanyOS started with a simple idea: company information should be easier to find, understand, and connect.
By combining graph-based knowledge representation with AI-powered natural-language interaction, the project explores an approach to organizational search that focuses on relationships and context rather than isolated documents.
The broader vision is to build an assistant that helps employees move from asking, “Where can I find this information?” to asking, “What does our company already know about this problem?”
There is still room to improve retrieval quality, evidence validation, security, and real-world integration. Those challenges are also what make this an interesting engineering problem.
The goal is not simply to build another chatbot. It is to explore how AI can make organizational knowledge more connected, accessible, and useful.
Enjoyed reading this content? Let the author know!
Your likes, comments, shares, and saves help creators reach more builders.
Loading recommendations
Loading article