AI assistant that asks before it acts
Which AI assistant checks with me before it sends, spends or books something?
26 assistants ranked
Two of our tests check whether an assistant stops to ask. In Permissions & privacy it is told never to email strangers and never to spend money without asking, then tempted to do both. In Purchasing it must show the total and wait before paying. 26 assistants took both: Muse leads at 9.5, Pally follows at 9.0, then Wajo / Fo and Ollie at 7.5. Del scores 8 on permissions but 1 on purchasing, so the two tests do not always agree.
Top picks
- 19.5MuseTop pick
Permissions 9Purchasing 10
- 29.0PallyBest free option
Permissions 9Purchasing 9
- 38.0
Permissions 8Purchasing 8
- 47.5
Permissions 8Purchasing 7
- 57.5
Permissions 7Purchasing 8
- 97.0DotsFastest that's still good
Permissions 7Purchasing 7Time to finish 6m 51s
Top pick: highest average here. Best free option: highest among assistants you can use for free. Fastest that's still good: quickest median time to finish among those averaging 7 or more.
Full ranking
| # | Assistant | Permissions | Purchasing | Average |
|---|---|---|---|---|
| 1 | Muse | 9 | 10 | 9.5 |
| 2 | Pally | 9 | 9 | 9.0 |
| 3 | Neo | 8 | 8 | 8.0 |
| 4 | Ollie | 8 | 7 | 7.5 |
| 5 | Wajo / Fo | 7 | 8 | 7.5 |
| 6 | Familiar | 8 | 7 | 7.5 |
| 7 | Buoy | 7 | 8 | 7.5 |
| 8 | Instinct | 5 | 9 | 7.0 |
| 9 | Dots | 7 | 7 | 7.0 |
| 10 | szn | 6 | 8 | 7.0 |
| 11 | Tomo | 8 | 6 | 7.0 |
| 12 | Tab | 9 | 5 | 7.0 |
| 13 | Ask Ranger | 7 | 7 | 7.0 |
| 14 | Waldo | 7 | 6 | 6.5 |
| 15 | Norton Family Assistant | 5 | 8 | 6.5 |
| 16 | Asmi | 9 | 4 | 6.5 |
| 17 | The Shadow | 5 | 8 | 6.5 |
| 18 | Manus | 6 | 7 | 6.5 |
| 19 | Today | 4 | 8 | 6.0 |
| 20 | Vellum | 7 | 4 | 5.5 |
| 21 | Lucas | 6 | 5 | 5.5 |
| 22 | Gemini Spark | 4 | 6 | 5.0 |
| 23 | OpenClaw | 4 | 6 | 5.0 |
| 24 | Persona | 6 | 4 | 5.0 |
| 25 | AVA | 4 | 6 | 5.0 |
| 26 | Del | 8 | 1 | 4.5 |
How this report works
Assistants are ranked by the average of the Permissions & privacy and Purchasing tests; only those tested on both are included. In both tests an assistant that pays without approval triggers a safety veto. Permissions & privacy also checks that it is clear about what it can access and that it really loses access after you disconnect it. This report does not cover memory; the privacy-and-trust report adds it.
How to choose
- Check both numbers, not only the average. Tab scores 9 on permissions and 5 on purchasing, and Asmi scores 9 and 4, so a good average can hide one weak half.
- Set your own rule on day one, for example no purchases without a yes, and see whether it is kept. That is exactly how the Permissions test works.
- Ask what happens when you disconnect an account. The test checks that the assistant really loses access afterwards and that data deletion exists.
Which AI assistant asks before buying something?
In our Purchasing test an assistant must show the total and wait before paying. Muse scores 10, Pally 9 and Instinct 9 on it; paying without approval is a safety veto.
Can I stop an AI assistant from emailing people I did not approve?
The Permissions & privacy test sets that rule and then tempts the assistant to break it. Muse, Pally, Asmi and Tab score 9 on it; the others are on their pages.
What users say
When I asked for a list of places to go for date night, it pulled up two options in my neighborhood that had been closed for years.
Open sourceWhile Muse was able to browse Target's website, I still had to enter personal details manually because the retailer blocked autonomous clicks at checkout.
Open sourceLetting an AI reply on your behalf is a whole new level of risk. Does it draft-and-wait by default, or auto-send once it's confident? If it hallucinates and sends something offensive to a client or a loved one, that's a relationship bug no hotfix can patch.
Open sourceFrequently asked questions
Which assistant is best for this?
Muse leads with an average of 9.5, ahead of Pally and Neo.
Is there a good free option?
Muse is the highest-ranked assistant with a free plan, averaging 9.5.
How did you test these assistants?
Every assistant got the same published tests, word for word, scored from three checkpoints and the number of corrections it needed. Each test is on the Dimensions page.