Skip to main content

​
What is the assistant?

The assistant answers your questions about Rapid Review and your account, in the Rapid Review app. Choose Ask AI at the top of an account’s page to open it. It opens in a panel on the right of the page, or over the whole screen on a phone. Open or closed, it stays as you left it as you move between pages. Type a question in the box at the foot of the panel, or pick one of the questions it suggests, such as “Why was my last push not reviewed?”. You can ask in your own language. The answers are written by AI, from Rapid Review’s documentation and what you can see of the account in the app. An answer can link the app’s pages it talks about, and propose changes for you to make with a button. The assistant never changes anything itself. An answer shows once it’s written and checked in full, which can take up to a minute. An answer that couldn’t be checked shows too, without change cards.

​
Who can use the assistant?

The assistant is on for every account. Everyone who can open an account in the app finds Ask AI at the top of its pages. Each person’s answers come from what that person can see in the app, so the account’s owner and a member can get different answers to the same question. If the assistant can’t answer just now, the panel says so. You can still reach a person on our team with Open a ticket.

​
What does the assistant know about my account?

What you can already see of the account in the app. The app reads it again for each question and sends it with the question:
  • the account, your role in it, and the page you asked from;
  • whether its reviews are running or paused, as the top of its pages shows;
  • the repositories you can see, with their review settings, when those last changed and whether someone in the account or Rapid Review’s operator changed them, and any application for free open-source reviews;
  • pushes of the last 14 days that weren’t reviewed in full, at most 30, each by its repository, pull request number and time, with why;
  • if you’re the account’s owner, its plan or trial, its credits left, and today’s credits against its daily budget, as on the Usage and billing tab;
  • the app’s pages you can open in the account.
The app never sends a pull request’s title, a commit, your repositories’ code, or another person’s GitHub login with a question. What you write in a question, such as a repository’s name or a piece of code, goes as part of the question. Each answer shows when it was written. Because the account is read again for each question, a later answer can give other figures than an earlier one, for example when reviews used credits in between.

​
Can the assistant change my settings?

No. It can propose a change, which you make yourself. A proposal shows under the answer as a card: the change in words, why, and a Make this change button. Nothing changes until you press it. An answer proposes at most three changes. It proposes only changes you could make yourself in the app:
  • As a repository’s admin: the repository’s review settings on the Repositories tab, such as whether it’s reviewed, its mode, how later pushes are reviewed and its severity floor; and applying for free open-source reviews for a public repository, or withdrawing an application.
  • As the account’s owner: the account’s severity floor, its daily budget, and its plan (below).
It never proposes the confidence tier, agent prompts, learnings, buying credits, ending a plan, invoices or payment details. For those, it points you to the page where you change them. Pressing Make this change does what the settings page does, with the same checks. If someone changed the same thing after the answer was shown, nothing is written, and the panel says so. A card whose change is already made says “Set already”, and a card you can no longer make, for example because you’re no longer the repository’s admin, isn’t shown. A change you make from a card is yours. The account’s Audit log names you, with “via the assistant” after your GitHub login, and so do the repository’s “Changed by” line on the Repositories tab and the daily budget’s “Last set by” line on the Usage and billing tab.

​
Can the assistant change my plan?

No, but if you’re the account’s owner, and you can pick a plan for it in the app, an answer can show plans to choose from. Each is a tile with the plan, how it’s billed, its price as Stripe has it and the credits it includes each month. Start or Switch to on a tile takes you to Stripe, where you confirm the plan and what it costs. Nothing is charged before you do. The assistant never ends a plan or buys credits. For those, it points you to the account’s Usage and billing tab (Plans and credits, Billing).

​
How are the assistant’s answers checked?

Every answer is written by AI and can be wrong. No person on our team reads it before you do. Under every answer, the panel says so: “Answers are written by AI and can be wrong. Open a ticket to reach a person.” Before you see an answer, another AI model checks it against our documentation and the facts about your account. It asks whether the answer states something neither supports, whether it promises anything, and whether it answers what you asked.
  • Where the check finds a problem, the answer is written again, more carefully, as far as the day’s limits allow, and checked again.
  • Where a doubt remains, a note under the answer says so, with Open a ticket.
  • An answer that seems to promise something, or that couldn’t be checked, comes without change cards.
Only a person on our team can promise a refund, credits, a date or an exception: the assistant isn’t meant to, and the check looks for answers that do. Questions about contracts, privacy, security and your personal data are for a person on our team too, so the assistant suggests a ticket instead. The only links in an answer are the documentation sections it used and the app’s pages it points to.

​
How do I reach a person from the assistant?

Choose Open a ticket under the conversation. The assistant writes a message to support from the conversation, as from you, and opens the support form with it. Read it and change what you like before you send it. Your message reaches a person on our team, who also sees the conversation it came from. Like any message from the app, it can carry facts about the account (Support mail and your data). You read the answer on the app’s Support page (How do I contact support from the app?). If the assistant can’t write the message, the form opens empty for you to write it.

​
What can I do with an answer?

Under each answer:
  • Copy puts the whole answer on your clipboard, as written.
  • Thumbs up or thumbs down tells us whether the answer helped. Press the same thumb again to take your vote back.
  • When it was written, such as “10 minutes ago”. Rest the pointer on it to see the exact time.
A file’s contents or a command, such as a .rapid-review.yml, shows in a code block of its own, with its language and a copy button for just that code. The copy buttons show where your browser lets the page use the clipboard.

​
How many questions can I ask the assistant?

Up to 30 questions an hour. The assistant also has a limit on how much it answers each day, for each person and for everyone together. When you reach a limit, the panel says so, and when you can ask again: a little later for the hour’s limit, and after midnight UTC for the day’s. A question that wasn’t answered stays in the panel, marked as not sent, with Send again. The assistant also writes only a few ticket messages for you an hour. Past that, write the ticket in your own words. Open a ticket reaches a person at any time.

​
Who sees my conversations with the assistant?

You do. The panel’s Your conversations lists your 20 latest conversations about the account, and New conversation starts another. No one else in the account sees your conversations in the app. A person on our team can read them, with the facts sent with each question, the changes proposed and your votes, to check and improve the assistant’s answers, and to help you when you open a ticket.

​
Who else receives my questions to the assistant?

Your question, the earlier turns of the conversation, the facts about your account and the changes you could make go, through the service OpenRouter, to two AI model providers:
  • Anthropic runs the model that writes the answers, and the ticket messages written from a conversation;
  • TypeSafe runs the model that checks each answer.
Neither provider may train its models on them. For checking, we also ask that the provider keeps nothing once it has answered. For answers, the provider may keep a request for a time under its own policy. These providers may handle your data outside the European Union, in particular in the USA. Where they do, the transfer rests on the EU–US Data Privacy Framework for providers certified under it, and otherwise on the European Commission’s standard contractual clauses.

​
How long are my conversations with the assistant kept?

A conversation, with your questions, the answers, the facts sent with them, your votes and the ticket messages written from it, is deleted 180 days after its last turn. If you opened a ticket from it, it’s kept until the conversation with support that the ticket went to is deleted too, 180 days after that one’s last message (How long is my support mail kept?). Your conversations with the assistant are tied to your GitHub account, not to an email address. To have them deleted sooner, ask from the app’s Support page while you are signed in: we delete them along with what you sent to support from the app. A request by mail deletes the support mail of the address it came from, once we can tell the request came from that address, and doesn’t reach your conversations (How do I ask you to delete my data?).