> ## Documentation Index
> Fetch the complete documentation index at: https://docs.youniverse.ai/llms.txt
> Use this file to discover all available pages before exploring further.
# Create and manage Agents
> Start an Agent, edit it, run it, test it, pause or stop it.
An **Aura Agent** is a teammate that works in the cloud while you don't. You describe the job in plain words; the Agent figures out the steps, the tools and the apps it needs each time it runs.
## Three ways to start one
Open **Agents** in the menu and choose **New Agent**, or tap one of the suggested ideas based on your connected apps.
Ask for something recurring ("every Monday…"). Aura offers **Start an Agent for this?** and opens it prefilled.
Paste a shared playbook link, or **Start a new Agent from this one**.
## Write a good job description
* **What** to do and **when** ("every weekday at 8 AM").
* **Where** to look and where to put results ("my Gmail inbox", "the Sales sheet").
* **Rules** ("never email customers directly", "cap spending at \$50").
* **How to report** ("send me a two-line summary").
See the [Prompt guide](/learn/prompt-guide) for templates.
## The Agent details page
| Area | What it shows |
| - | - |
| **Header** | Name, short handle (e.g. AG-7K3P), trust level and status |
| **Last run** | When it last ran and its next scheduled run |
| **After every run** | Rules to follow each time a real run finishes (tests don't count) |
| **Playbook** | Reuse or share what the Agent learned — see [Memory and playbooks](/agents/memory) |
| **Tabs** | Workflow, Memory, History, Files, Vault, Triggers, Alerts, Browser |
## Run, test, pause, stop
* **Run now** — starts a real run immediately.
* **Test** — a safe rehearsal. Tests never skip approvals and don't trigger "after every run" rules.
* **Pause** (yellow) — stops new runs but keeps pending questions waiting for you.
* **Stop** (red) — switches the Agent off until you **Reactivate** it.
* A pulsing green dot means it's working right now.
## Adaptive workflow
The **Workflow** tab says *Adaptive workflow — the exact actions can change each run*. Agents don't follow a frozen step list: they choose tools at run time, so they keep working when an app or page changes. What worked last time is saved as a recipe to make the next run faster.
## Talk to your Agent
Every Agent has its own chat. Message it mid-run ("skip the weekend emails") and it will adjust and remember the instruction. It asks you questions there when it needs something.
## Reliability built in
* Long jobs keep going past time limits automatically.
* If a run stalls for five minutes it's diagnosed and retried once; a second stall stops it and alerts you.
* Schedules use your own time zone.
* Real-world actions are never repeated when a run resumes.
Pro and Elite.
Runs use credits based on the AI work and tools used. You can set spending limits per month and per run — see [Trust and spending](/agents/trust).
This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.