Zarvis is OpenZcash's guide to Zcash, and this page explains its chat by showing it: the badges under answers, the questions you can try, what a complete answer contains and how the depth choice changes it. Limits, retention and the incident log live on the About page.
Under every answer a badge says which kind you got. These samples are rendered by the same code the chat uses, with the model configured right now:
Direct from OpenZcash data · no AI
The question matched one of the site's own records and was answered by code: no model call, no cost, and the record's date is shown. Trust it as you trust the site's data pages, and check the date.
Claude Opus 5.5 · 3 sources · check them
The badge names the model that wrote the answer and counts the sources it used. Open the source list and check them. A numeric guard checks USD, ZEC and percentage figures against the retrieved records and withholds figures it cannot verify, but other answers and currencies are not fully covered, and a matching figure can still be out of context.
Each link opens Zarvis with the composer filled in. A question link fills the composer with that one question and leaves the address bar; nothing is sent until you press send.
Answered by site code from the cached quote, with its time and a stale flag. No model call, and the quota unit is returned.
Answered from the site's own treasury records, with the data dates shown.
Answered from the official ZIP corpus at a pinned revision: title, status, dates, abstract.
Answered from the reviewed protocol FAQ; the same text every visitor gets.
Answered by the configured model from retrieved records; the badge names it and counts the sources.
A writing request: the model drafts it from the reviewed pitch library.
A sample model answer, annotated. The numbers mark what to check on every real answer:
Q: How is the Dev Fund split, and what changes at the next halving?
The block subsidy is split 80% to miners and 20% to the Dev Fund. At the next halving the subsidy itself halves, so every one of those shares scales down together. What the Dev Fund pays for today, and what the rules say about what follows, is on this site's funding pages.
The selector next to the composer sets how much thinking an answer gets. Auto, Quick and Deep analysis are the three choices; these budgets come from the same planner the chat runs:
| Quick | Deep analysis | |
|---|---|---|
| Answer written within | 55 seconds | 110 seconds |
| Reasoning effort | low | high |
| Research rounds over the records | 3 | 4 |
| Reply-length ceiling | 3,000 tokens | 6,000 tokens |
Auto, the default, adapts to the question: analytical requests such as compare, analyse, trade-offs or sensitivity switch to Deep analysis on their own. The selector next to the composer always wins when you choose.