How do I get started with Cortex?

Connect the agent you already use to your Socra Account and continue your real work. With Cortex instructions and the required permissions, the agent retrieves relevant knowledge and saves useful decisions as it works, including the first Module in an empty Account. You do not need to write a knowledge library before starting. Cortex stores knowledge in Modules; Flash returns a Module and the other Modules it needs.

Connect one agent

First ask your agent to list the Cortex Modules it can access. A real, successful tool call confirms access, even if the list is empty. An answer written without a Cortex call does not verify the connection.

If Cortex is not connected, use the installation guide to choose the method supported by your agent: a plugin, a remote MCP connection, or the Socra CLI. During authorization, select the Account that owns your knowledge. Signing in to the Cortex website does not connect a separate agent.

For CLI access, install the Socra CLI and its Account and Cortex plugins before running these commands. Sign in and check the selected Account:

socra account login
socra account status
socra cortex module list

Complete sign-in in the browser and wait for the terminal to confirm success. A successful list verifies that CLI connection; it does not verify a separate MCP connection. If you already have a working connection to the intended Account, keep using it.

Start from your work

Tell the agent what you need to do. Flash known targets directly; otherwise prefer keyword Search. With the CLI, replace your topic with the subject:

socra cortex module search --query "your topic"

Use socra cortex map only when Search leaves subjects or relationships unclear.

Once the relevant Module is known, replace module-name with its actual name and Flash it:

socra cortex flash module-name

The Flash result includes the requested Module and its Dependencies, with prerequisites first. Ask the agent to read the complete result and explain which decisions apply to the task. It must still check the current code, running system, and other evidence needed for the work.

An empty Account is valid. Continue your real task without creating sample knowledge merely to finish setup. When useful knowledge emerges, the agent can save it in the appropriate Module, briefly name the affected Modules, and explain what changed and why.

Decide how the agent should maintain knowledge

Tool access and instructions are separate. The suggested Cortex instructions tell an agent when to retrieve knowledge and when to save or update useful decisions during normal work. Supported plugin skills or hooks deliver that guidance to the agent; a connection alone does not establish that it received the instructions. You can adapt the instructions, including asking for your approval before changes. The agent host's tool-confirmation controls also affect when it can act.

If you add persistent instructions, verify in a new conversation that the agent both received the guidance and can make a Cortex call. Do not treat a successful connection as proof that instructions were saved. You can also give instructions as you work.

The browser app at cortex.socra.com lets you inspect your Account's Modules and work. Use the same Account there and in your agent.