Some jobs are not a single question. Read these ten contracts, pull out the renewal dates, compare them and draft a summary. An AuraDoc agent takes a goal written in plain language, breaks it into steps, works through your documents and reports back. You stay in control: you can see an estimate before it starts, ask for approval on risky steps, and review every action afterwards.
Start with a goal
You describe the goal in up to 4,000 characters and choose the documents it should work on, one or several. A run can be synchronous, so you wait for the result, or asynchronous, so it works in the background while you do something else. You can limit the number of steps, from 1 to 12, with six as the default, and you can pick a model preset: fast, balanced or powerful, or name a specific model. A finished run can be repeated, continued from where it stopped, or cancelled while it is still working.
Templates and skills
Typing the same goal again and again is tedious, so agents come with templates grouped by category. Templates marked as skills have a defined input and output, which makes them predictable to run. You can also create your own templates with a name, a category and an icon, and attach autonomy settings to them. Your team can then launch the same trusted agent with one click instead of rewriting the instructions each time.
See the cost before you run
Before launching, you can ask for an estimate. It returns a suggested model with the reason for the suggestion, the complexity of the goal, an estimated number of tokens, an estimated cost and an estimated duration, plus the models available to you. It is a forecast, not a promise, but it helps you decide whether a broad goal should be split up, or whether a lighter model is enough. Statistics for finished runs, such as success rate, average duration and most used templates, are available too.
Approval before risky steps
An agent that can act should not act unchecked. When you turn on approval for a run, it pauses at each step and waits: you approve the step to continue, or reject it to stop that path. Agents have extra guardrails on top. By default they are limited to 10 actions and a token and cost budget per run, and certain actions, such as sending an e-mail or a WhatsApp message, deleting something, applying a workflow, requesting a signature or writing to cloud storage, always go to a person first. Actions the agent is not confident about, below a confidence threshold, also wait for approval, and pending approvals expire after 48 hours by default. A dry-run mode lets you see what an agent would do without doing it.
Let agents start on their own
Agents do not have to wait for you. A trigger starts one automatically when something happens: a document arrives in a folder, a WhatsApp message or an e-mail is received, a document of a given type appears, a schedule fires, or a webhook is called. You can test a trigger before enabling it, see which documents it has handled, and keep a separate schedule of recurring agent runs with a goal, documents and an interval. A control room shows what is running, what is waiting for approval and what has failed.
Audit, sharing and an emergency stop
Every run keeps an audit trail of the events that happened, and you can export a run for your records. If you need a colleague or an outside reviewer to see a result, you can share a run through a link that expires after the number of hours you choose, from 1 up to 720, with 72 as the default. Agents can also keep memories about your documents, which you can list and delete whenever you wish. And if something looks wrong, an administrator can flip a single kill switch that stops agents for the whole workspace.
Where to find it
Agents live in their own section of the dashboard, with runs, templates, triggers, approvals and the control room. If you want a more conversational way to work with a document, see AI chat with citations. To connect agents with your other systems, read about webhooks and the REST API, and to keep control over who may do what, see team roles and permissions. You can create an account and try an agent on your own documents.