Skip to content

Cortex overview

View as Markdown

Cortex helps you stop repeating yourself to AI. Your agents preserve useful preferences, requirements, decisions and reasons while you work, then retrieve that knowledge when it matters in another task or conversation.

You do not need to write a knowledge base first. Connect your agent, give it the Cortex guidance, and keep doing the work you already do. Explain an idea, review a result or correct a decision. The agent maintains the knowledge and its connections alongside that work.

Start with your agent

Open guided setup (new tab) and choose Ask my agent to help, or give your agent the installation guide. You complete sign-in and required consent; the agent handles the setup it can perform and checks the connection and guidance.

After setup, continue a real task. A new workspace can start with an empty graph. Useful knowledge accumulates and improves later work as agents preserve and reuse it.

How knowledge grows

The plugin teaches the agent when to find existing knowledge, what specifics are worth keeping, and how to update them. The agent searches before creating anything, preserves reasons and dependencies, names the affected Modules and explains what changed and why. When later work reveals a correction, it updates the relevant knowledge instead of leaving another conflicting copy.

You can allow this maintenance or ask the agent to recommend changes and wait for approval. The custom instructions make that choice explicit. Host permissions and workspace access still apply.

Existing documents, code reviews or other material you authorize the agent to read can provide a starting point. You do not need to reformat them into Modules. How agents grow your knowledge explains extraction, changes, team sharing and the limits of what an agent can learn.

How Cortex works

Knowledge lives in versioned Modules, each stating what is defined for a subject and why. Dependencies connect the knowledge needed to understand it. A Flash assembles a Module and its dependencies for the agent to read before acting.

Issues track work to reconcile code and other artifacts with that knowledge. Saving a Module does not change the code; the agent carries out and verifies that work separately.

Who controls the knowledge

Cortex shares knowledge with people and CoAgents through Module permissions. A decision saved during one person's work can guide another authorized teammate without copying it into each tool. Workspace membership alone does not grant access to every Module.

Knowledge never moves between workspaces. This security boundary keeps each team's knowledge isolated. Authorized agents can retrieve and maintain knowledge within the selected workspace across tools and sessions. The host controls tool confirmations and execution; Cortex enforces data access. See Access and permissions.

Cortex receives what you and your authorized tools send to it. It does not automatically import every conversation, inspect your repository, monitor an expert's work or learn unshared knowledge. The agent does the capture and maintenance during the work it can see and perform.

Ways to use Cortex compares the agent integrations, web app, CLI and API.