Set up Cortex with your agent

Use an agent you already have. Cortex supports Codex, Claude Code, Cursor and GitHub Copilot CLI; the integration guide covers those and other hosts. You do not need a Socra AI teammate for this exercise.

As of October 31, 2026, Coagent costs $20 per person per month, including Cortex, Channels, Tasks, Drive and Admin and $20 in monthly usage credit, with no trial. Your AI subscription is separate. Connecting an agent on your own computer adds no AI teammate charge. Check current pricing before subscribing.

Paste this into your agent:

The agent follows the maintained instructions and checks the connection. If you prefer to follow the steps yourself, Cortex setup offers the supported paths. A successful call, including an empty list, confirms access; opening the setup page does not.

Part of setup is teaching the agent how to use what we’ve just connected.

Cortex provides instructions for your agent. They explain when to retrieve knowledge, how to preserve what it learns from you, and how to keep that knowledge connected and current. Supported coding-agent plugins include this guidance; other connections need it saved in the agent’s persistent instructions.

Ask your agent to adapt the Cortex instructions to how you work and save your preference for future conversations, preserving any instructions you already use.

I want my agents to maintain my knowledge as we work, without asking me to approve every Module edit. My direction is:

You might prefer to review every change first. You can tell your agent:

The agent still does the writing and figures out where the knowledge belongs. You choose how much review you want. It can help save that preference alongside the Cortex instructions, so you don’t have to explain it in every conversation.

Once setup is finished, try it with one correction that matters to you. You might keep asking an agent to put decisions before background in your weekly update. Give it a real update to work on and explain why the order matters: people need to see what needs their attention before reading the history.

Ask the agent to save that preference and its reason in a Module, reusing the existing owner when one fits. Have it connect the Module to the reporting guidance the work needs, including a teammate’s existing Module when it has access. These Dependencies bring the required knowledge together when the agent retrieves it. Ask it to show you the saved decision and give you the Module name.

If the update does not yet follow the decision, have the agent create an Issue for the change and its check. It can then Flash the relevant Module, revise the update and show you whether the decisions come first without losing the background you needed. Have it record the checked result and finish the Issue, correcting the Module if your review changed the decision. You now have a useful draft and knowledge another connected agent can retrieve.

Keep the saved knowledge appropriate for its audience. For this example, the reusable writing preference may belong in a team Module; a private personnel matter from the update does not. Ask the agent to check who can access that Module before putting sensitive details there. The access guide explains how sharing works.

Now open a fresh conversation with a connected agent. Give it the Module name and a new update to write, without repeating the preference. Ask it to Flash the Module with its Dependencies and use that knowledge. Read the result: are the decisions first, and did it keep the background you still needed?

If the preference wasn’t retrieved, fix the connection or retrieval step. If it was retrieved but misunderstood, correct the saved knowledge. If it was understood but ignored, show the agent the mismatch and have it repair the work. Those are different problems; another paragraph of instructions won’t fix all of them.

I use this with our design system because I want a correction to guide the next agent. One Growing Brain explains how those decisions connect across our products.

Look at the second result before adding more knowledge: did the agent apply your saved decision correctly without you teaching it again? Keep working and correcting what needs your judgment. The first-assignment guide can help you give your agent another useful responsibility.