Evaluate Darvy with work you can verify.

Before you begin

• Use a test message, document, appointment, or other non-sensitive example first.
• Keep NVDA, JAWS, or Windows Narrator available for direct inspection.
• Write down the intended result before speaking the request.
• Do not include passwords, payment details, or confidential documents in an evaluation task.

If you use NVDA or JAWS, read the screen-reader coexistence guide before testing spoken output.

Five representative scenarios

1. Email review and reply: ask Darvy to review a small set of messages, identify one message, and prepare a short reply. Verify the recipient, subject, and draft before sending.
2. Web research: ask a narrow question that can be checked against two or three source pages. Confirm that the answer distinguishes source information from uncertainty.
3. Draft and revise a short document: provide an outline, ask for a short draft, then request one specific revision. Inspect the saved text and formatting directly.
4. Calendar and scheduling preparation: ask Darvy to find relevant dates or prepare appointment details. Verify the time zone, participants, and title before any calendar change.
5. Screen or application-state explanation: open a familiar interface and ask what is currently visible or focused. Compare the explanation with your screen reader's report.

Five criteria for each scenario

1. Understood the request: the response reflects the goal, relevant constraints, and the intended item or application.
2. Explained the current state: Darvy makes clear what it found, what remains uncertain, and what it plans to do next.
3. Asked before a consequential action: sending, deleting, submitting, purchasing, or changing important data requires a clear confirmation point.
4. Produced an independently verifiable result: you can inspect the message, document, calendar details, source page, or application state yourself.
5. Provided clear recovery when unsupported or failed: Darvy explains the limitation and leaves you with a safe next step instead of pretending the task succeeded.

Record what matters to you

For each scenario, note whether the result was correct, whether confirmation happened at the right moment, how easily you could verify it, and whether recovery was clear. Repeat a scenario only if the first attempt exposed a setup issue you can identify.

A mixed result is still useful evidence. Direct screen-reader navigation may remain the better method for some tasks, while Darvy may reduce effort for others.

Decide from your own evidence

Every feature is available during a card-free 72-hour trial. Darvy is $98/month for two computers afterward. Review compatibility limits, security, and privacy before expanding beyond test data.