# One Growing Brain for All Your AI Agents

There are many agent harnesses, and you probably use many of them, but there is only one Cortex for all your AI agents.
So far, we’re the only team who have solved this problem:
How do I transfer to an AI agent what it needs to know before it acts?
If you’re a senior engineer watching your systems degrade in quality with the addition of AI agents, Cortex was built with you in mind. At every step, we chose the decision that would be best for you.
That's because Cortex was built by another senior engineer, Mike Morton,  who really needed this solution himself, because my agent swarm was wrecking his life's work.
For the past year, he has been solving this problem. We named the technical solution SemDAG (Semantic Directed Acyclic Graph).
Now he can share his Modules with me, only the ones he wants to share, so my agents can flash his knowledge (in milliseconds!) before acting. They learn what they need to know to do things right. My swarm no longer erodes what makes Mike’s work special.
What Cortex helped us build in seven weeks
Agent execution: Our agents completed more than 3,500 tracked Issues in roughly seven weeks because Cortex gave each agent the governing decisions, dependencies, ownership, and work history before it acted.
My technical reach: I can create repositories, provision databases, build services, fix production systems, and deploy applications without knowing how to do those things myself because Cortex gives my agents the technical and architectural knowledge contributed by Mike, even when he is unavailable. My agents used that knowledge to build eduarda.ai in just a few minutes.
Design system: I created a global design system, visible at socra.design, that governs our current products and is designed to support hundreds more. I state a design decision once, and my agents preserve it in Cortex and apply it consistently across the code.
Autonomous software factory: I created Fab, which pulls Issues from Cortex, flashes the required knowledge, assigns work, verifies results, and continues through the remaining inventory. I had tried to build it for a long time, but it only worked once Cortex gave every agent the same knowledge and durable work state.
Socra Cloud: Mike built Accounts, projects, OAuth, app identities, licensing, billing, package publishing, builds, workloads, Sandboxes, secrets, and deployment paths by storing their boundaries and relationships in Cortex for agents to follow.
Marketing pages: Mike and I developed the product positioning and story together, while Cortex gave every agent the same voice, design rules, and approved claims across all our marketing sites.
Documentation: We built consistent product documentation because Cortex gave each Agent the current product behavior, terminology, audience, and technical rules before they wrote.
Production mail system: Our agents built verified domains, mailboxes, shared ingress, and native SMTP delivery because Cortex gave them the service boundaries, security rules, and infrastructure knowledge needed to work safely.
Support: We built support for people and agents because Cortex stores each approved answer as a versioned Module and publishes the same source to the website and agent-readable exports.
These just to name a few. We even took a two-week road trip traveling to Colorado, all the way to Glenwood Springs, and Austin.
Cortex allows our a small team to move incredibly fast because very decision we make compounds. We continue to accelerate our improvement as every piece of specific knowledge our own AI agents build into our own Cortex compounds.
Cortex feeds these pieces of knowledge we call Modules to your agents so they learn what they need, when they need it.
Install the plugin, that's it
Plugin Installation Guide: Connect a Coding Agent
All you have to do is install the plugin, and your agents learn how to use Cortex. You can even forget it exists.
You can continue to use your agents the way you do. You will notice having to repeat yourself less and the quality of your AI agents' output increasing. It starts slow and continues to get better as your knowledge compounds.
Now it’s available via plugin for Claude Code, Codex, Cursor, and Copilot. Also available via MCP, CLI, which you can use to connect any agent.
If you’re using MCP or CLI, add these instructions to your claude.md, agents.md - they will teach your agents what Cortex is and how to use it. You don't need to do this if you're using the plugin. You can also copy and paste this:
Cortex augments any agent with the owner-governed knowledge it needs before it acts and coordinates work through Issues on the Module Semantic DAG (SemDAG), where Modules are the single source of truth upstream of code and agents create and update them through owner-approved changes so people never have to repeat what they know.

- Modules contain owner-governed knowledge about how something should work and why.
- Dependencies compose Modules into a knowledge graph.
- Flash loads a named Module and its complete dependency closure into your context before you act.
- Issues coordinate bounded work that reconciles Modules (what should be) with repositories (what is implemented).
- Inbox shows the unblocked Issues you are responsible for now.
Use Cortex to:

- Flash the narrowest relevant Module before acting on conversation or work involving specific knowledge or decisions, including architecture, constraints, priorities, or work history. Find it with context_map or context_search. Treat flashed knowledge as governed intent; verify code and external reality separately.
- Coordinate work through Issues and Inbox. Read each Issue and flash its owning Module before acting. Attach new Issues to the narrowest Module that fully governs the work.
- Refine the SemDAG when conversation or work reveals specific knowledge that could improve future decisions, conflicts with a Module, reveals outdated or incomplete knowledge, or requires a missing Module or Dependency. Flash the narrowest relevant Module first. If it already makes the same decision, do not restate it. If not, propose the exact update to its owning Module, or a new Module with Dependencies if none owns it. Keep durable should-state, not incidents, status, implementation, migrations, or cleanup. Request explicit owner approval under # Module Change Request, with the exact change and before and after for updates. Apply only after approval, then flash the changed Module again.
Try Cortex in one repository with one decision you are tired of repeating. Let your agent flash that knowledge before it works, then watch the magic.
If something could be better, tell us. You are early enough to help shape what Cortex becomes.