Start using Cortex in your work

You do not need to fill Cortex with knowledge before using it. Connect your agent, then keep working. The agent preserves useful decisions, preferences and reasons as they emerge, so later work can build on them.

Connect your agent once

Give your agent the installation guide, or open guided setup (new tab) and choose Ask my agent to help. It handles the setup it can perform; you complete sign-in, Account selection and any permissions your host requires.

Setup checks two things: the agent can actually call Cortex, and it receives the guidance that teaches it when to retrieve and maintain knowledge. An empty Module list is normal for a new Account. You do not need a separate CLI connection if your chosen agent integration already works.

Continue a real task

Ask for the work you already wanted done. Explain a decision, correct a result, or share an artifact the agent may use. You do not need to name a Module or design a graph.

For example, while reviewing an interface, you might say:

Keep selected text readable in both themes. Use the same treatment throughout the product.

That correction gives the agent work to do and knowledge worth preserving. With Cortex guidance and the required access, the agent:

1. Finds the existing design knowledge and reads it before changing anything.
2. Preserves the current requirement and its reason in the relevant Module, creating one only if needed. It maintains the dependencies that make the requirement understandable.
3. Verifies what it saved and briefly explains the change.
4. Carries out the requested work. If code or another artifact needs to catch up with the defined system, an Issue tracks that reconciliation.

You keep reviewing the result as you normally would. The agent handles the knowledge maintenance alongside the work.

Choose how much to review

The supplied guidance tells agents to maintain useful knowledge. You can instead ask for recommendations before anything is saved:

Recommend changes to Cortex knowledge, but show me the proposed changes and wait for my approval before saving them.

Or allow maintenance within the material and access you have authorized:

Keep our Cortex knowledge current as we work, and tell me what changed and why.

Ask your agent to preserve this preference in its persistent instructions. Host tool confirmations and Cortex access controls still apply. Knowledge maintenance does not grant permission to publish, deploy or take unrelated actions; see the security model.

Let later work use what came before

On a later relevant task, the agent can Flash the design Module and its dependencies before making a decision. Another authorized agent connected to the same Account can use the same knowledge. If you change the requirement, tell your agent in the course of the work; it updates the existing knowledge instead of leaving a conflicting copy behind.

A new graph starts small. Its value grows as useful knowledge is preserved and reused. The useful result is a later task that needs less repeated explanation and applies the decision correctly, not a particular number of Modules.

Already have useful material?

Let your agent use relevant material it is authorized to read: a design review, existing documentation, code with explanatory comments, or a conversation you provide. It can identify durable decisions, compare them with existing Modules and ask about conflicts. You do not need to rewrite those sources into Module format.

For that workflow, and for keeping knowledge current across a team, read How agents grow your knowledge.