> ## Documentation Index > Fetch the complete documentation index at: https://docs.atako.ai/llms.txt > Use this file to discover all available pages before exploring further. # Creating and managing agents > Every section of the agent creation sheet, plus agent settings, lifecycle, licenses, and permissions. An agent is a persistent, always-on AI worker — not a one-off chat session. Creating one takes a minute; the choices you make below shape how it behaves for its entire lifetime. ## Who can create agents Company admins can always create agents. Other members need the **Can create agents** permission, granted per member by an admin from **Settings → Users**. If your company has no active subscription, or all its active agent licenses are taken, agent creation is blocked. See [Subscription](/guides/subscription). ## Starting a new agent On the **Floor** storey, click **Create an agent**, or press **N**. The agent then follows your cursor: drop it on a team to join that team, or off to the side to create a new team (creating a team is admin-only). The creation sheet opens next. You can also open the sheet directly at `/floor?newAgent=1`. ## The creation sheet A **Name** — shown everywhere the agent appears, including to the people who chat with it — and its **Team**. Pick **Individual agent** to leave it without a team. Three fields, all editable later in the agent's settings: * **Mission** — 1 to 3 sentences: the outcome expected from this agent. * **Instructions** — method, tone, and limits. They override its default style, never the security and confidentiality rules. * **Business context** — facts specific to this agent's role. Be concrete about scope: what the agent owns, what it should never do without asking, and how it should sound. Vague instructions ("be helpful") do less than a short list of do's and don'ts. Every agent already receives your company's shared context, set once in the **Knowledge** cube on the **Context** storey. Use the agent's own business context only for what's specific to its role. See [Team](/guides/team#shared-company-context). The harness decides what the agent can do. Two are offered: | Harness | For | | - | - | | **Hermes** | A general-purpose agent: integrations, channels, scheduled tasks. | | **opencode** | A coding agent: GitHub repositories, one session per piece of work, draft pull request. | An **opencode** agent needs an active GitHub connection for your company and at least one **Allowed repository**, in `owner/repo` form. It replaces the **Integrations** section below. See [Engines](/concepts/engines). The harness is final. It can't be changed after the agent is created. The integrations your company has already connected, each with **Read** and/or **Write** access for this agent. Write implies read. An integration left without permission is not connected to the agent. You can change these at any time in the agent's settings. See [Integrations overview](/integrations/overview). **Atako inference** needs no key and is billed with the rest of your usage — see [Atako inference](/concepts/atako-inference). **Your own key** sends model calls from your own provider account, if [BYOK](/concepts/credits#byok-agents) is enabled for your company; those agents never draw from your shared credit balance. You can switch the inference source later in the agent's settings. Pick a model from the catalog, or leave the default. Search by name, star models to keep them under **Favorites**, and find the ones you picked before under **Already used**. **Suggested by Atako** lists a short selection; **All models** lists the full catalog. The model is final. It can't be changed after the agent is created. Click **Create agent**. The agent uses the time zone of the member who created it, as set in their **Settings → Appearance** — its scheduled tasks run in that zone. There is no company-wide time zone. The creator's time zone is read again whenever the agent is resumed or updated. Channels aren't part of the creation sheet — you turn them on afterward. See [Channels overview](/channels/overview). ## Agent settings Open an agent on the Floor, then click **Agent settings** in the agent panel. Only the agent's creator, or a company admin, can change its settings. Each change is saved on its own and applies immediately. | Section | What you can change | | - | - | | **Identity** | Name, team, and job title | | **Brief** | Mission, instructions, and business context | | **Integrations** | Read and write access per company connection — or, for an opencode agent, its allowed repositories | | **Runtime** | The inference source. Model and harness are shown read-only. | | **Lifecycle** | Resume, retry setup, export, or update the agent, depending on its state | | **Danger zone** | Delete the agent | ## Lifecycle | Status | Meaning | | - | - | | **Preparing** | The agent's environment is being set up. | | **Active** | The agent is running and uses one of your company's active agent licenses. | | **Unresponsive** | The agent isn't responding right now. Your messages are delivered as soon as it's back. | | **Paused** | Compute is stopped, but everything the agent has built up — memory, scheduled tasks, files — is kept. Resuming restarts the same agent, not a new one. | | **Failed** | Setup failed. Retry it from the agent's settings. | | **Completed** | The agent has ended. Its active agent license is freed. | Agents are paused automatically in some cases — for example, when a free trial ends without a subscription in place — and can be resumed from **Lifecycle** once the underlying issue is resolved. Deleting an agent is permanent: its work stops and its memory disappears. The conversation and exchanged files remain available. ## Active agent licenses Every active agent uses one active agent license from your company's pool. Licenses aren't tied to human users: your team can have any number of members regardless of how many licenses you own. See [Subscription](/guides/subscription). ## Related The concept behind agents and sub-agents. Hermes and opencode compared. Working with an agent day to day. Roles, permissions, and inviting teammates. This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.