AI agents for personal assistant

Handle email, tasks and errands for one person from natural language.

18 agents, each with its inputs, outputs and review level checked by a person.

Personal assistant agents handle email, tasks and errands for one person from natural language: drafting replies, keeping a to-do list, booking things, reminding you and answering questions over your own files. They sit closest to the person and so need a high level of trust.

A newer group of always-on personal agents goes further. Each runs on its own computer in the cloud, with a browser and sometimes a desktop, and keeps working when you close the app: it can research, fill in forms, compare options, draft documents and come back with a result or a question hours later. Many of these are in beta, on a waitlist or invite-only, and their features change quickly.

The listings record what each agent connects to, what it does without asking, what it drafts for confirmation, which accounts it acts in and how it is accessed (chat, mobile, desktop, voice). Choose by how much it should do alone. Start with tasks where a mistake is cheap, keep approval on for anything that sends, buys or deletes, and check the review level on each listing to see what is confirmed with you.

The agents

Groups that share agents with this one.

Questions

What is an always-on personal agent with its own computer?

It is an assistant that runs in a cloud machine of its own, with a browser and tools, rather than only answering in a chat window. You give it a task, it works through the steps on that machine, and it reports back or asks for a decision. You can usually watch or take over the session. The listing records how each vendor describes it.

Will it ask before sending an email or making a purchase?

Most ask by default before anything with consequences, such as sending a message as you, paying or deleting. Some let you relax this for named contacts or spending limits. The review level on each listing shows the vendor's default; keep approval on until you have seen how it behaves.

Which logins and accounts does it use?

Some connect through authorised integrations to your email, calendar and files; agents with their own computer may ask you to sign in to sites inside their browser, and keep that session. Prefer connections you can revoke, avoid sharing passwords in chat, and check the inputs on each listing for what the vendor documents.

Is what I share with it kept private?

That depends on the vendor's terms: how long conversations, files and screenshots are kept, whether staff can review them and whether they are used to train models. The vendor's privacy page, linked from the listing, describes this; the catalogue does not verify it. Leave sensitive documents out until you have read it.

Can I use one now, or is it invite-only?

Many of the newer agents are in beta, on a waitlist or open only to some plans or regions. The listing records availability as the vendor stated it on the date it was checked, which can change quickly.

Is there a free tier?

Several assistants publish a free tier with limits on tasks or usage; each listing shows the published pricing with the date it was checked.