You can create your own AI agents in Tana just by describing what you want. No setup, no prompt engineering.
An agent is an assistant you delegate to. You equip it with skills (reusable instructions) and integrations (connected tools like GitHub, Slack, and Linear), and it works for you on demand, in a meeting, or on a schedule while you sleep. And because agents and skills are shared, building a good one is something your whole team gets, not just you.
What an agent is
An agent is an assistant you configure for a job and then talk to. It has its own prompt, its capabilities, an optional voice, and an optional schedule. Converse with it in chat or by voice, or set it to run on its own.
Keep an agent to yourself, share it with a teammate, or open it to the whole org. A capability one person builds becomes the team's with nothing extra to set up, and a useful agent becomes a shared teammate.
What you equip it with
An agent's capabilities decide what it can do, in two parts:
- Skills are reusable instructions for specific jobs: "file this as a GitHub issue," "summarize this meeting in our format," "draft a PRD from these notes." A skill does the same job the same way every time, for everyone who uses it, so your whole team files bugs or drafts PRDs in one consistent shape. You can also run a skill on its own, with
/in chat, from an item's Send to menu, or from the Skills button in a meeting. - Integrations connect the agent to your tools, like GitHub, Slack, and Linear, so it can act in them rather than just talk about them.
You toggle both on in the agent's Customize panel; together they set what the agent can read, create, and reach.
Build a meeting-prep agent
You do not configure agents by hand, you describe them. Here is the whole flow, building one agent that preps you before every meeting, the job Walk into every meeting prepared puts on autopilot.
Click the pencil button at the top of the sidebar, or ask in chat: "Create an agent that preps me for every meeting by pulling context on the people I'm meeting with and what we discussed last time." Tana scaffolds the agent from that one description: its prompt, a name, and a starting set of capabilities.
Open the agent's Customize panel and give it what the job needs. Turn on a skill that writes the prep in your format ("summarize our last meeting and list open items per attendee"), and an integration when the context lives in another tool, like Slack or your CRM. Together, skills and integrations set what the agent can read, create, and reach.
Not quite right? Tell it: "also include any open tasks assigned to the people in the meeting." You refine an agent the same way you created it, in plain language, never by editing settings by hand. See Agents for the full reference.
Now choose when it runs:
- Chat or voice, when you want to ask it for a prep on demand.
- On a schedule, so it runs while you sleep and tomorrow's preps are waiting in your Library each morning.
- As a space's default assistant, picking up new chats started in your team's space.
Because the agent is a doc, you can keep this one to yourself or share it so the whole team gets the same prep. The more of your repeatable work you hand to agents, the more Tana does on its own, so you spend your time on the conversation, not the busywork.