Docs

What the AI assistant can do

The in-app "Ask AI" assistant looks up your candidates, forms, and metrics, screens candidates with AI review, drafts emails, builds forms, moves candidates through stages, and runs the custom skills you save - with examples of what to ask.

What it is

Binary has an in-app AI assistant you open with Ask AI. Ask it in plain language and it looks things up, screens candidates, drafts messages, and makes changes for you. It works within your role and plan, so it won't surface or change anything you can't reach yourself.

While it works, an indicator above your message names the step it's on - "Creating the job…" rather than a plain "Thinking…". It's a live status, not a log: once the reply lands the indicator is gone, and reopening the chat later shows only the reply itself.

Look things up

  • Candidates - find someone by name or email and see their answers, stage, labels, ratings, and files. It can also read an uploaded résumé and pull their LinkedIn profile.
  • Forms - descriptions, questions, stages, and submission counts.
  • Interviews and feedback - a candidate's interviews and the feedback interviewers left.
  • History - a candidate's stage timeline: which stage, when, and who moved them.
  • Team - who's in the workspace, and their role.

Report on your hiring

Ask for counts, rates, and trends - submissions, review and response rates, interviews, feedback, referrals, page views - per form, per recruiter, or over time. You can also ask for the people behind a number, so "12 interviewed" becomes the list of who.

  • Average time to first review, grouped by recruiter, is attributed to whoever reviewed the candidate first. It measures the candidate's wait, not that reviewer's working speed. Say "candidates waited about X hours," never "so-and-so takes X hours to review."
  • Average rating, grouped by recruiter, averages the ratings that person gave in interview feedback. It is a different number from the overall per-submission average rating, since raters disagree. Never present the two as the same metric.

Average rating grouped by recruiter can't be combined with another filter in the same query; if that query errors, drop the extra filter rather than retrying the same shape. Time to first review has no such restriction. Recruiter breakdowns only include people who have access to the page, so a candidate's own application never appears as a stage move someone made.

Screen candidates with AI review

  • Set up AI review on a form - describe what you're looking for and the assistant turns it into criteria, then grades candidates Yes / No / Maybe against each one with a short justification. See AI review.
  • Choose the depth: a fast review reads the application only; a deep review also pulls the candidate's LinkedIn profile for background the application doesn't cover.
  • Review new candidates, which is the default, or all of them.
  • Ask about the results - who matched everything, who missed a specific criterion, how many matched each one.
  • Turn on auto-review to have every new applicant graded as they apply. This needs a paid plan.

Get things done

  • Move candidates to a stage - one person or a whole filtered group, such as everyone who matched all your criteria. This is internal; candidates aren't notified.
  • Draft emails to one or many candidates, personalised per person. The assistant writes the draft and hands you a Send button.
  • Build a form - it creates the form as a draft, invisible to candidates, can write the description and questions, and can reuse the style of a form you've already published.
  • Edit or publish a form - change questions and settings, then publish, pause, close, or archive it.
  • Manage who can see a form - list, add, or remove its editors.
  • Workspace basics, for admins - edit your public page's description, or rename the workspace.
  • Try Binary with sample data - it can create a demo form pre-filled with example candidates.

Anything your team or your candidates would see - publishing a form, editing a live one, changing your public page, granting access - is confirmed with you first. The assistant tells you what will change and waits for your yes.

Open pages without losing the chat

When the assistant points you to something, it shows an Open page card. Click it and Binary takes you there.

On desktop the assistant stays open beside the page you land on, in a split view you can resize by dragging the divider, so you can keep asking follow-ups without losing your place. It reopens each time the assistant sends you somewhere new; choose Ask AI in the sidebar to bring it back if you closed it. On mobile, opening a page closes the chat and takes you there full-screen - reopen Ask AI to pick the conversation back up.

A plain link written inside a reply behaves by destination: one pointing into Binary navigates on click, while one pointing anywhere else opens a small popover offering Copy and Open in new tab, so a stray click can't take someone out of the app. Only external links do this.

Give it reusable skills

A skill is a saved instruction you write once and reuse - a shortcut for a request you make often, like "write a rejection email in our tone" or "summarize this candidate against our leveling rubric."

To use one, click the + button beside the message box and pick it from the list, or type / and start typing its name. The skill applies to that one message: the assistant follows its instructions for that reply, on top of whatever else you ask in the same message.

To create, edit, or delete a skill, choose Manage skills from that same menu. Each skill has a Short name you'll use to find it, the Instructions the assistant should follow, and an optional When to use note that helps the assistant recognise on its own when the skill applies.

Answer how-to questions

Ask "how do I…", "where is…", or "can Binary do…" and the assistant answers from this documentation, and can link you straight to the right screen.

Things you can ask

  • "How many candidates applied this week?" · "What's the review rate for the Designer role?"
  • "Who applied to the Engineer form?" · "Who did Sarah refer?"
  • "Show me Jane's application" · "How did their interview go?"
  • "Screen everyone for 5+ years of React and based in Bengaluru."
  • "Who matched all my criteria? Move them to Shortlisted."
  • "Draft a rejection email to everyone I archived."
  • "Create a form for a backend engineer, based on our last one."

Notes & limits

  • Candidate search matches name or email only - not company, title, skill, school, or location. To narrow people down by those, ask for an AI review with that as a criterion, or use labels and filters on the Responses page. See Search and filter candidates.
  • The assistant can't add labels, edit a candidate's answers, delete candidates, delete a form, or schedule interviews. It also can't choose which jobs are pinned on your public page.
  • It never sends an email by itself - you always press Send.
  • Where your plan or role can't reach the data, it says so rather than guessing, and offers an upgrade prompt where the plan is the reason.

Also in Slack

You can ask the same lookup and analytics questions from Slack, on any plan - see Connect Slack. Slack is read-only: screening, messaging, and editing forms happen in the app.

On this page