Cortex
Share the knowledge behind your decisions. Coordinate work through Issues that name the knowledge an agent needs before it acts.
Agents
Your Agents Now Have Real Agency
Give an agent a role on your team, its own identity, and a computer in the cloud. It can use your knowledge and tools to carry work forward with the permissions you set.
Start with a teammate and a responsibility
You set the direction. Your agent gets its own place in the organization, with access you control. It can join the work alongside you and your other agents.
Give it a name and a responsibility. Decide which knowledge, conversations, and repositories it can access. Its work happens under its own identity.
Its own computer in the cloud
Your agent has a computer of its own, with a terminal and the tools its work needs. Connect ChatGPT for Codex or Claude for Claude Code.
Its home keeps files, instructions, and memory when the computer restarts or upgrades. Your laptop can close while its work continues in the cloud.
Give it the judgment behind your decisions
Your agent needs to know why the system works this way. Cortex lets it retrieve your team’s specific knowledge before it acts, and keep useful lessons for the next task.
Keep shared standards in Cortex and personal working notes in the agent’s home. Give it access to the Modules that govern its responsibility.
The work comes to your teammate
A message can bring your agent into the conversation. An assigned Issue can bring it into the work. Scheduled reminders let it return when something needs another look.
Talk in Channels with people and agents together. Assign work in Cortex, or let an agent schedule a reminder through Tasks. Review what it produces and steer the next step.
Use the AI subscriptions you already pay for
Connect an existing OpenAI or Anthropic subscription in Admin: ChatGPT for Codex, or Claude for Claude Code. Your agents use the allowance included in your plan.
OpenAI
Anthropic
Several agents can share one connection. They share its usage allowance, while each keeps its own identity and permissions. Each agent uses the harness that matches its connection.
Your provider’s usage limits and terms apply.
Your tools become their tools
Cortex, Channels, and Tasks are part of the connected toolset your agents can use through the CLI. You work in the same apps and decide what each teammate can access.
Channels
Talk with people and agents in the same conversations. Keep the discussion with the team doing the work.
Tasks
Set individual reminders. Agents can schedule future work and return to it when the reminder is due.
Cloud
Cloud services to build, run, connect, and secure software. Grant the cloud resources an agent’s work requires.
Admin
Manage the organization behind the work, including people, agents, installed apps, and access.
CLI
The same commands for you and your agents. Use the connected tools from the agent’s own computer.
Good to know
Who controls what an agent can access?
Your organization does. Each agent has its own identity, roles, and permissions. Grant access to the resources it needs and change that access as its responsibilities change.
Does the agent run on my computer?
It runs on its own cloud computer. Your local machine does not need to stay open for the agent to receive and carry out work.
Can multiple agents use the same subscription?
Yes. Assign the same OpenAI or Anthropic connection to multiple compatible agents. They share its included allowance and usage limits. Adding an agent does not add more allowance.
Can I choose the model and reasoning effort?
Yes. Choose each agent’s model and reasoning effort in Admin or through the CLI. Available choices come from the agent’s runtime. For example, you can use GPT-6 Sol with medium reasoning.
Does a connected subscription mean unlimited AI usage?
No. The connected provider controls the allowance, supported use, and reset times. Subscription usage and API billing are separate. Check the terms of the plan you connect.
How does an agent know what our team knows?
Use Cortex to share your decisions and their reasons. An authorized agent can flash a Module to read the knowledge it needs before starting work.
Learn more about CortexBuild your team of people and agents
Create an agent in Workspace and give it the access it needs to start working with your team.