Agent, Team, and Company Memory: Three Scopes That Change Everything

Most AI platforms treat memory like a monolith: one bucket of context, shared nowhere, lost at the end of a conversation. Cortex takes a different approach. Memory in Cortex operates across three distinct scopes: agent, team, and company. Each scope serves a different purpose, and knowledge flows bidirectionally across them. This architecture transforms how AI agents learn and collaborate.

The Three Scopes Explained

Agent Scope: Personal Knowledge

Agent memory is individual to each AI agent. This is where the agent accumulates the specific context that makes it effective at its particular role.

Agent-level memory includes:

• Working style preferences: How this agent approaches problems, whether it prefers detailed explanations or concise summaries, how it prioritizes requests
• Project-specific context: Accumulated knowledge about ongoing projects, client quirks, stakeholder preferences, and historical decisions
• Interaction patterns: The agent remembers previous conversations with users, their communication style, what questions they typically ask, and how the agent has successfully helped them before
• Task specialization: When an agent handles the same type of work repeatedly, it builds a personalized knowledge base about that domain

Imagine a customer support agent that handles a specific product. Over time, the agent accumulates knowledge about the most common support issues, workarounds discovered through conversations, and how to recognize when an issue requires escalation. This knowledge is unique to that agent and makes it more effective every day.

Team Scope: Shared Workflows

Team memory sits between individual agents and company-wide policies. This is where teams document how they work together and manage specific client relationships or projects.

Team-level memory includes:

• Client assignments and context: Which clients work with which team members, client preferences, communication protocols, specific requirements
• Domain conventions: Industry standards, terminology, best practices specific to the team's field
• Runbooks and workflows: Step-by-step procedures for common processes; documented patterns that work for this team
• Shared decision history: Why certain technical choices were made, what was tried and failed, institutional knowledge about the team's approach

Consider a financial services team using Cortex. They maintain a team memory that includes templates for regulatory compliance documentation, client relationships mapped to specific agents, and domain-specific terminology. When a new team member joins, they inherit this knowledge immediately.

Company Scope: Organizational Backbone

Company memory is the foundation. This is where organizations document policies, principles, and decisions that should influence all AI agents across the company.

Company-level memory includes:

• Organizational policies: Privacy standards, data handling procedures, approved vendors and tools, security protocols
• Glossary and terminology: How the company refers to products, internal jargon, proper names and acronyms, brand voice guidelines
• Architecture decisions: Why certain technical systems were chosen, constraints and limitations of current infrastructure, future roadmap
• Org structure and contacts: Team hierarchies, who owns which functions, escalation paths, relevant contact information

A healthcare organization using Cortex would maintain company memory that includes HIPAA compliance guidelines, approved patient communication templates, proper medical terminology, and the organizational hierarchy for handling sensitive requests.

How Knowledge Flows: Inheritance Down, Graduation Up

The three-tier memory system is not siloed. Knowledge flows in both directions, creating a dynamic learning system.

Inheritance: Knowledge Flows Down

When an agent operates, it can access memory from team and company scopes. This is inheritance: policies and context automatically available to all agents within a team or company.

If the company memory states "all customer communication must include an unsubscribe option," every agent is aware of this requirement through inheritance. If a team documents that "client X prefers email communication on Tuesdays," agents assigned to that team inherit this context.

Graduation: Knowledge Flows Up

As agents work, they accumulate insights that may be valuable beyond their individual scope. Cortex's graduation system identifies when agent-level knowledge should be promoted to team or company scope.

The graduation mechanism uses convergence detection: when multiple agents independently arrive at the same insight (compared at cosine similarity >= 0.90), the system recognizes this as validated knowledge. No LLM evaluation needed. The insight graduates automatically from agent scope to team or company scope.

Example: A customer support agent discovers a workaround for a recurring technical issue. Another agent independently documents the same workaround. When convergence detection identifies both agents have reached the same conclusion, the workaround is automatically promoted to team memory, where all agents can access it.

Alternatively, signal-based promotion uses LLM validation to identify particularly valuable single-source facts that warrant promotion to higher scopes. A security insight or a significant process improvement might be promoted this way.

Practical Example: A Product Team

Let's trace how the three scopes work together in a real scenario.

A product company uses Cortex to manage customer success. The company scope contains the product glossary, customer privacy policy, and organizational structure. The customer success team scope contains client relationship mappings, common troubleshooting procedures, and team-specific workflows.

Agent Alice is assigned to a specific customer. Agent-level memory stores Alice's interaction history with that customer, preferred communication style, and specific issues the customer frequently encounters.

When customer service requests come in, Alice draws on all three scopes: she knows company policies (company scope), understands team workflows and the customer's background (team scope), and recalls previous interactions and personalized context (agent scope).

Over time, Alice discovers that this particular customer often needs a custom integration guide. She documents it in her agent memory. Meanwhile, another agent working with a different customer discovers the same pattern. When Cortex detects convergence, the insight graduates to team memory, so all customer success agents can benefit.

Eventually, if this pattern spreads across multiple customer success teams, it graduates again to company scope, where the knowledge is integrated into the product glossary and official documentation.

The Competitive Advantage

Traditional AI systems offer no organizational scoping. They're designed for individual conversations, not learning across teams or companies. Cortex's three-scope architecture creates something different: a learning organization where insights compound, policies are enforced consistently, and knowledge grows more valuable every day.

The three scopes transform AI from a tool you use (and then forget about) into infrastructure that learns from every interaction and makes every agent in your organization smarter.

Experience three-scope memory in action. Sign up at launchcortex.ai and deploy agents that learn and grow across your entire organization.