# Set up Cortex with your agent

When I correct an agent, I want the next one to use what we learned. You can try this with one of your own corrections: connect an agent to Cortex, let it save the decision and check the next piece of work in a fresh conversation.

## Connect your agent

Start with an agent you already use. Cortex supports Codex, Claude Code, Cursor and GitHub Copilot CLI; the [integration guide](https://workspace.socra.com/docs/cortex/integrations) lists the supported connections. You can use an agent on your own computer for this exercise.

As of November 12, 2026, [Socra HQ costs $20 per person per month](https://socra.com/pricing), including Cortex, Channels, Tasks, Drive and Admin and $20 of monthly usage, with no trial. Your AI subscription is separate. Connecting an agent on your own computer adds no AI teammate charge.

Paste this into your agent:

> Connect Cortex by following https://workspace.socra.com/docs/cortex/installation.md. Handle the setup you can and guide me through sign-in or consent that needs me. Verify the connection with a real Cortex call, then check whether a fresh conversation receives the Cortex instructions. Tell me what is connected and what remains to do.

The agent handles setup and checks access. You handle sign-in and consent. If you prefer to follow the steps yourself, use [Cortex setup](https://cortex.socra.com/setup).

## Choose how your agent keeps knowledge

The connection lets your agent reach Cortex. It also needs [instructions for using it](https://workspace.socra.com/docs/cortex/instructions.md): when to retrieve knowledge and how to keep what it learns from you. Supported coding-agent plugins include this guidance; other connections need it in the host’s persistent instructions.

I want my agents to maintain my knowledge while we work. I give them this direction:

> As we work, keep what you learn from my ideas, decisions and corrections in Cortex. Create and update the relevant Modules, maintain their dependencies and use that knowledge before acting. Keep it current when things change, and tell me what you changed and why.

You can choose to review changes before they’re saved instead. Ask your agent to preserve your preference alongside the Cortex instructions without replacing your existing guidance. The agent does the writing and decides where the knowledge belongs.

## Try one correction on real work

Use a correction you already care about. For example, I might ask an agent to put decisions before background in a weekly update, because readers need to see what needs their attention. Give it a real update and explain your preference and the reason.

Ask it to save that decision in the relevant Module, the document that holds the guidance for this work. If the decision needs other knowledge to make sense, the agent connects those Modules as Dependencies. Have it give you the Module name when it’s saved.

Then have it revise the update and check the result. If there’s unfinished work to coordinate, it can track that in an Issue. You read the draft and correct anything that still misses your intent.

Keep the reusable preference separate from private details in the update. The agent should check the Module’s audience before saving sensitive information. The [access guide](https://workspace.socra.com/docs/cortex/administration/access-and-permissions) explains who can see shared knowledge.

## Check it in a fresh conversation

Open a fresh conversation with a connected agent. Give it the Module name and a new update, without repeating your preference. Ask it to Flash the Module, which retrieves the guidance and its Dependencies, and use it on the new assignment.

Read the result. Did the decisions come first, with the background you still needed? If not, ask the agent to inspect what it retrieved. Missing knowledge needs a connection or retrieval fix. An unclear decision needs better wording. A decision it understood but ignored needs a correction to the work.

This is the small version of what I use across our products. [How I Use Cortex](https://eduarda.ai/cortex) tells the story behind it. Keep using your judgment on the work that comes back; that is where the next useful correction comes from.

Canonical page: https://eduarda.ai/how-my-agents-know-what-i-know

Flash bundle: https://eduarda.ai/how-my-agents-know-what-i-know/flash.md
