About Pagent
An independent scorecard for the AI assistants you can text.
Every assistant promises to run your errands by text, and every demo shows a different task, so they're hard to compare. We wrote the tasks down, published them, and run the same ones against every assistant we can get into.
How we test
- Our testers use dedicated accounts set up the same way for every assistant, with the same inbox, calendar and contacts.
- Every score comes from a logged run of a published test.
- Untested means blank. We never estimate a score.
- Bookings and orders made during tests are cancelled or returned. Phone-call tests only ever call our own numbers.
Independence
Vendors can't pay to be tested, ranked or reviewed. Anyone can request a test, and requests from vendors are labelled as such.
Something wrong?
If you think a score is wrong, or a product has changed since we tested it, request a retest. The tests are public so they can be argued with. See how scoring works.
Use our data
The whole leaderboard, every score included, is at /data/leaderboard.json, and a plain-text summary for AI assistants is at /llms.txt. You're welcome to cite it: link to the page you used and give its update date.
Contact
To get an assistant tested, use the request form or email [email protected].