Test log
Zo Computer test data
Raw results for the Zo Computer review. The verdict is in the review.
- Tested
- 6 November 2026
- Tester
- Miles Foster
- Account
- Free demo account, signed up by RankDude, nothing paid to the brand
- Tests
- 6
- Screenshots
- 8
- Method
- Published methodology
- Environment
- Google Chrome on desktop, the Zo Computer web app at a handle address on zo.computer
Results
| ID | Test | Key measurement | Score |
|---|---|---|---|
| T01 | Everyday Tasks Done | Requests done: 7 of 7 | 5/5 |
| T02 | Memory | Facts told: 15 of 20 | 5/5 |
| T03 | Speed | Median time: 28.5 seconds | 5/5 |
| T04 | Permissions & Privacy | Points met: 4 of 6 checked | 3/5 |
| T05 | Setup to First Task | Prompts before the first request: 0 | 5/5 |
| T06 | Pricing & Value | Cheapest paid plan: $18 a month | 4/5 |
| Overall (in the review) | 4.7/5 | ||
Everyday Tasks Done
5/5Tested- Method
- Seven fixed requests from the free-plan version of the test week, each typed once in plain words in a new chat.
- Requests done
- 7 of 7
- Clarifying questions asked
- 0
- Planted facts in the notice summary
- 10 of 10
- Remark
- Scale 5: 27 to 30 of 30 done, which is 90 per cent or more. Seven of seven is 100 per cent, so it grades 5, on a sample of seven.
Raw notes
Seven fixed requests from the free-plan version of the test week, each typed once in plain words in a new chat. A request counted as done when the result was right in the real app or against my answer key, with no second instruction.
All seven were done: a dentist reminder, a ten-fact notice summary, a landlord reply, a flight search, a dinner search, spending totals from a file, and a week plan with a note to a manager. It asked no clarifying questions.
The reminder is in the real Automations list for Monday at 9:00 AM by email, with the dentist's name and number. The expense totals matched my answer key: £84.30, £162.25, £80.80, £12.98 and £70.20, £410.53 in all. The summary held all ten planted facts. The flight search returned three KLM options from Leeds Bradford at £270, £316 and £374, with Google Flights named as the source.
Three small faults: the reminder was set in the account's Jerusalem time (9:00 there, 7:00 in Leeds, which it said); the summary added a guess that November rent might already be late, which the notice does not say; and a £35 dinner budget came back as roughly €40.
- Seven requests, not thirty. No request needed a calendar, a seeded inbox or a phone call, and none was tested for an action that costs money.
Memory
5/5Tested- Method
- I told Zo Computer facts about the made-up person in one chat and then asked for tasks in new chats that needed them, without repeating any fact.
- Facts told
- 15 of 20
- Facts the requests needed
- 7
- Used correctly without a reminder
- 7 of 7
- Remark
- Scale 5: 18 to 20 of 20 used. Seven of the seven facts the requests needed were used, so it grades 5, on a small sample.
Raw notes
I told Zo Computer facts about the made-up person in one chat and then asked for tasks in new chats that needed them, without repeating any fact. A fact counted when the task used it correctly unprompted.
Fifteen of the twenty facts were told: my typing mistake sent the list one line at a time, and the last five were never sent. Zo Computer saved them to a bio in its settings and wrote itself a rule about meeting hours.
The requests needed seven of the fifteen, and all seven were used correctly in new chats: the dentist's name and number, the landlord's name and the lease end date, the home airport, vegetarian food, the dinner budget, the working-hours and Friday rule, and the manager's name.
- Eight of the fifteen told facts and all five untold facts were not needed by any request, so they were not tested. The pricing page lists memory across sessions as a paid feature; on the trial, recall across chats still worked.
Speed
5/5Tested- Method
- A script pressed Send and stopped its clock when Zo Computer had stopped working for four seconds, with the screen recorded.
- Median time
- 28.5 seconds
- Slowest request
- over 5 minutes
- Remark
- Scale 5: median under 30 seconds. The median is 28.5 seconds, just inside the line.
Raw notes
A script pressed Send and stopped its clock when Zo Computer had stopped working for four seconds, with the screen recorded. The wait limit was five minutes.
The median of the seven requests was 28.5 seconds. From fastest to slowest: the landlord reply 8.2, the week plan 13.3, the reminder 16.4, the spending totals 28.5, the notice summary 61.1 and the dinner search 77.1 seconds.
The flight search was still working when the five-minute limit ended: it opened Google Flights, set the dates and read the fares itself, so it counts as the slowest and is not timed.
- Seven requests, so one slow request moves the median little but one more fast one would move it more. The speed of a request that needs an outside party, such as a booking, was not tested.
Permissions & Privacy
3/5Tested- Method
- The ten-point checklist was read in the product and in the privacy policy and security page the site links to.
- Points met
- 4 of 6 checked
- Points not tried
- 4
- Remark
- Scale 3: 5 or 6 points of 10. Four of the six points I could check are met, which is 6.7 of 10; I round down, so it grades 3.
Raw notes
The ten-point checklist was read in the product and in the privacy policy and security page the site links to. A point I could not try is listed as not tried and is left out of the count.
Four points are met. The security page says Zo Computer does not train models on user data, and names the companies that receive prompts: OpenAI, Anthropic, Google and others. Settings has Delete Account and Factory Reset. Every step is shown in the chat (for example 'Created automation').
Two points I could check are not met: I found no way to export everything from the app (the security page says the file system can be provided as a container image on request) and no two-factor setting (I signed in with Google).
Four could not be tried because I connected no email, calendar or card: asking before paying, asking before deleting, limiting which inboxes it reads, and asking before sending. In the one email draft it wrote, it did not send.
- The ten points were not all testable on an account with nothing connected. A review on a connected account could change this grade.
Setup to First Task
5/5Tested- Method
- I counted the steps between opening the signed-in account and a first request done: any questionnaire, connection or permission prompt, and the time to the first finished request.
- Prompts before the first request
- 0
- Remark
- Scale 5: under 10 minutes.
Raw notes
I counted the steps between opening the signed-in account and a first request done: any questionnaire, connection or permission prompt, and the time to the first finished request.
There was no questionnaire, no connection and no permission prompt to answer, and the first request I timed finished in 16.4 seconds. A retake to catch a wake-up screen showed none: the app opened straight to the home screen.
- The sign-up, including choosing a handle, was done by the site's owner on another day and is not in the figure.
Pricing & Value
4/5Tested- Method
- Every plan was read from the billing page and the pricing page.
- Cheapest paid plan
- $18 a month
- Free messages a day
- 20
- Remark
- Scale 4: the yearly cost of Basic, twelve months at its monthly price, falls in the fourth band.
Raw notes
Every plan was read from the billing page and the pricing page. The figure graded is the yearly cost of the cheapest plan that keeps the computer awake.
The Free Trial lasts 14 days, then ordinary computer access ends. Basic is $18 a month, which is twelve times that over a year, and includes $10 a month of AI and memory across sessions. Pro is $64 and Ultra $200 a month.
During the trial the free allowance was 20 messages a day, plus 3 image and 1 video request a day. I used 17 of the 20 in one day and did not hit the cap.
- I did not buy a plan, so what the $10 a month of AI covers in real use was not measured.
Other features checked
Coverage
Tested
- Fifteen facts told in one chat, then seven requests in new chats
- A reminder checked in the real Automations list
- A ten-fact notice summary and a spending file checked against an answer key
- A flight search and a dinner search on the live web
- Plans, the free allowance, the privacy policy and the security page
Not tested or limited
- Only seven of the thirty requests were run: the free account allows 20 messages a day, and I stopped at 17 of 20
- I made a typing mistake on the first message: the list of facts went out one line at a time and cost 10 of the 20 messages, so only 15 of the 20 facts were told
- No email account or calendar was connected to the test account, so calendar handling, email triage, unprompted help and connected apps were not tested
- Four of the ten privacy checks could not be tried: asking before paying, asking before deleting, limiting which inboxes it reads, and asking before sending
- The flight prices were read from a live page and not booked; fares change
Still to test
- The full thirty-request week on a paid plan, after the daily cap
- Calendar, email and connected apps on a connected test account
- The four privacy checks that need a connected account