Rainbook

AI

Ask questions about your pipeline, read summaries and health reasons, draft follow-ups and approve agent proposals, all inside Jira, with each person seeing exactly what Jira lets them see.

What Rainbook's AI reads

Rainbook's AI runs on models hosted by Atlassian. It reads what you can see in Jira and nothing more. Your data isn't used to train models.

That is the whole rule. When you ask a question, open a summary or draft an email, Rainbook's AI works from every record, email and file Jira lets you open, and from nothing else. Your colleague asking the same question gets an answer built from what they can open. Nobody has to set up a separate AI permission model, and nothing your team can already see is held back from the AI that helps them.

Every number Rainbook's AI gives you counts every record that belongs in it, or says in words what it leaves out: "Open pipeline is $1.2M. This excludes 3 deals you can't open." Totals never quietly drop records.

Rainbook's own AI runs on Forge LLM, which Atlassian hosts. Your requests are processed in Atlassian's AI processing locations, and nothing goes to any other model provider.

Rainbook's AI answers in your language: the one you use Jira in, or the one Rainbook shows you. It answers in English, German, Spanish, Japanese, Brazilian Portuguese and Simplified Chinese; if you use Jira in European Portuguese you get Brazilian Portuguese, and in any other language, English. Ask summaries, answers about an account and briefs only you can read come in your language, and you can ask in any language. Email drafts follow the language of the conversation, or yours when the conversation doesn't make it clear. What Rainbook writes into Jira stays in English, so everyone on the site reads the same values: record summaries and next steps, handoff briefs, comments, proposed changes and the meeting briefs your team shares.

In each of these languages, Rainbook checks the numbers, percentages and dates in an AI summary or brief against your records and doesn't show one that states a figure they don't hold; an email draft highlights such a figure for you to check. Its check for invented names is narrower outside English: a one-word name in German, a name that opens a sentence, and a name written in Japanese, Chinese or Korean script are not always caught, so read names in those answers with care.

Rainbook's sensitive-data checks that look for words read English words only: the words that mark a social security, passport, A-number or medical record number, ICD diagnosis codes, the health, adverse-event and criminal-history word lists, age phrases such as "years old", the titles Mr, Mrs and Ms, and CUI and ITAR markings. Card numbers, US social security numbers written as 123-45-6789 and IBANs are found in any language.

Ask

Ask for a report the way you'd ask a colleague: "Which renewals are at risk this quarter?" or "Win rate by owner, last two quarters". Ask turns it into a chart in a few seconds and writes a short summary under it.

  • See how it was built. Every answer shows the metric definitions, filters, dates, grouping and data freshness ("Data as of 10:42; Jira changes take up to 15 minutes to appear"), worked out from the question, never invented by the model.
  • Edit the chips. Each part of the answer (metric, group by, filters, period, comparison, chart type) is a chip you can change directly, with no new question needed.
  • Follow up. Add "only Enterprise" or "compare to last year" and Ask refines the same chart. A thread holds 20 follow-ups on Standard and 40 on Advanced.
  • Save it. Pin a chart to your Home, keep it in Reports, or put it on a Jira dashboard. Saved reports refresh for each viewer under their own permissions and use the same certified CRM metrics every time.
  • Share it. Choose Everyone using Rainbook as the shared scope. Share copies a link. A colleague who opens it sees your questions with results re-run on the records they can open, each with its own total note.
  • Ask about one account. On any record, the right rail answers questions such as "What did they say about pricing?" from the emails and attachments you can open, quoting the sentence each claim rests on and linking the thread by its subject. When it reads only part of a long history, it says so: "Based on the latest 40 of 212 emails".

Ask history

Your Ask threads are listed in Ask history and kept for 13 months from the last time you used them. You can delete any thread you created, or choose Clear my history to remove all of them at once. Drafts you create are kept the same way and cleared with your history.

Summaries and next steps

Every Deal, Account and Renewal carries a short AI summary and up to three next steps, each with an owner ("us" or "them"), a due date where one was mentioned and a link to the email it came from. Captured email threads get their own summary too.

Summaries refresh when something changes: a new email on a linked thread, a stage or close date change, or a Meeting that ends. Click Refresh summary any time you want a fresh one. Opening a record never waits on the AI, so pages stay fast.

Every summary shows "Updated" with its age and an "AI summary" label. When it was built from part of a long thread, it says which part ("From the first and latest 20 of 40 messages"). Rate it with the thumbs control so Rainbook keeps getting better for your team.

Summaries and next steps need a paid plan or a trial.

Health and its reasons

Every account and deal has a Health score from 0 to 100: Healthy at 70 and above, Watch from 40 to 69 and At risk below 40. The score comes from plain rules (support volume, SLA breaches, open bugs, reply lag, stage age, days to renewal and the signals your industry pack adds), so it is the same every time and works in JQL: Health < 40.

The three biggest factors behind each score become its reasons. On paid plans Rainbook's AI writes them as one or two plain sentences; on the free tier they are clear templated sentences. Either way you can see why an account turned At risk without opening a report.

Drafted follow-ups

On a record or an email thread, click Draft follow-up. Rainbook reads the thread, the attachments you can open (CSV, XLSX, PDF and DOCX), your open deals and next steps, and writes a reply you can edit. Add an instruction such as "propose Tuesday" or "shorter" and regenerate.

Use Copy, Open in Gmail or Open in Outlook to send it from your own mailbox. Rainbook never sends email for you. Drafts in the Gmail and Outlook sidebars work the same way; see Sidebars.

Drafts need a paid plan or a trial.

Duplicates and merge

Rainbook spots likely duplicate accounts, contacts and deals as they are created, during an import dry run and every night. Clear matches appear in your Inbox as "Possible duplicate" and as a banner on both records; closer calls appear on the records only. On paid plans, Rainbook's AI checks each close call and settles it where it can. Only people who can open both records see the suggestion.

From the banner, choose Merge, Not a duplicate or Open both. Merge shows a field-by-field comparison so you pick what the surviving record keeps. When you can edit both records, your click applies it; otherwise the record's owner gets an Approval request. After a merge you see "Merged · Undo" for 10 seconds, and Undo merge stays in both records' activity history for 30 days.

Proposals you approve, with Undo

Rainbook's AI never changes a record on its own. Every change it suggests is a proposal, and a person applies it.

  • Tell the right rail what to do. On a record, type "move close date to 15 Jan and add a follow-up for Tuesday". When you can edit the record, the change applies at once and the side sheet shows exactly what changed.
  • Capture suggestions. When an email says the deal moved ("we'll sign on the 20th"), Rainbook proposes the update to the deal owner, quoting the email.
  • Renewal save plays. For a renewal due within 90 days on an account below Health 40, Rainbook proposes a save play: follow-ups, a comment and a next step for the renewal's owner.
  • Promises. When a customer email carries a commitment ("EU data residency by 15 January"), Rainbook proposes tracking it as a promise, linked to the deal and to the engineering work that delivers it.
  • Shipped for you. When a Jira release ships work a customer asked for, each account owner gets a drafted customer note and a proposed next step in their Inbox.

When you can edit everything a proposal touches, your own click applies it, with no extra dialog. When someone else has to approve it, Rainbook creates a Jira Approval request for them, so approvals live in Jira with everything else.

After any applied proposal you see "Applied · Undo" for 10 seconds, and the record's activity history keeps an Undo on that entry while the changed fields still hold the applied values. Fields someone changed since are left alone and listed as "Changed since, not undone". Admins who want two people on every change can turn on Require a second approver in Settings › AI and Rovo on any paid edition; it is off by default.

Rainbook agents

Six Rainbook agent roles (Capture, Research, Follow-up, Deal desk, Renewal and Handoff) work in the open on your board, in Atlassian's AMP Patterns. An agent can own a pipeline column, so a deal entering that column starts its work, and every result arrives as a proposal a person accepts. Delegate to Rainbook sends a research brief and a drafted customer note to each account owner, marked "delegated by" the person who asked. The Agents settings page shows each role's acceptance rate and lets an admin pause any role or stop them all.

Show sources

Every AI result has Show sources. It lists the records and emails the result used, each linked, filtered to what you can open. The person who asked can also see the full prompt and output, which makes it easy to check an answer and to report one that missed. A Rainbook admin can open it too, with any record or email they can't open shown as withheld. The person who asked sees an item withheld in the same way once they can no longer open it, and the full prompt is hidden whenever anything is withheld, because it mixes every record and email together.

For each AI result, Rainbook keeps the prompt, the sources it read and the output for 30 days in your own site's storage, for Show sources and your feedback, and then deletes them. They never go to logs, they leave your site only in a diagnostics bundle an admin chooses to send to support, and they are never used to train or tune a model. Customer data never trains a model.

The Rainbook agent in Rovo

Ask Rovo about your pipeline. Each person sees exactly what Jira lets them see.

In Rovo Chat, the Rainbook agent answers questions about your customers, deals and pipeline: "What's in my pipeline this quarter?", "Summarise Acme", "Which renewals are at risk?". It finds accounts, summarises an account or deal, runs any Rainbook report, gathers the context for a follow-up email, shows an account's work across sales, promises, engineering and support, and proposes updates. It names every record with its key, such as "Acme Corp (RBDIR-12)", and quotes numbers exactly as Rainbook returns them, with the note on anything they leave out.

When you ask it to change something, it proposes the change. If you can edit the record, it gives you a link to apply it in Rainbook with one click; otherwise it tells you which Approval request it created and who approves it.

Two Rovo skills come with it, Weekly revenue-and-delivery review and Deal review. The same playbooks for Claude Code are coming soon, with AI clients.

You can also build your own agents in Rovo Studio that use Rainbook's tools, under the same rules: each person's agent sees what that person can open, and changes wait for a person.

Rainbook's own AI is included in your subscription and runs on Atlassian-hosted models; the Rainbook agent in Rovo uses your Rovo plan. The agent works on Jira plans that include Rovo.

AI clients: Claude, ChatGPT and others, coming soon

Claude, ChatGPT and other AI clients are coming soon. Connected to Rainbook's tools through Atlassian's MCP server, each will get the same answers the Rainbook agent gives that person, email summaries and quotes included. Until then, ask the Rainbook agent in Rovo.

When AI clients arrive, a change one asks for through Rainbook's tools follows the same rules as the Rainbook agent. Atlassian's own MCP tools, a separate Atlassian service, can edit Jira work items directly; those are ordinary Jira edits under the person's permissions.

Your own email reaches your AI clients once you agree. Turn on Use my email in AI clients in Settings › AI and your email. One agreement covers the Rainbook agent in Rovo Chat and Rovo Studio agents now, and clients such as Claude or ChatGPT when they arrive, and you can turn it off any time. Summaries of other people's email reach AI clients, for records each person can open, only when an admin turns on Admin → AI → Include other people's email summaries for AI clients; it is off by default. This use is named in the text people see before they connect a mailbox, and it is never used to train a model.

Rovo and the Teamwork Graph

In plain Rovo Chat, ask what is at stake with a customer before a renewal, and Rovo answers with the cited deal, requests and promise from the Teamwork Graph. Accounts and deals your team shares go into the graph with Jira's permissions; private records and private deals stay out, and so does email text.

In Rainbook Beta now, and in Rainbook from January.

The AI budget meter

Rainbook's own AI is included in your subscription. Admins see it as AI actions in Settings › AI budget:

  • AI actions used and available this month, and the reset date: "3,125 AI actions a month on Standard. 1,940 used. Resets 1 Nov."
  • Two totals, People (Ask, drafts, proposals) and Automatic (email and record summaries, health reasons), with the split by feature and a daily chart. Mail Sync's drafts are counted too, with a note saying when that figure was last updated.
  • Personal limits: the number of Ask analyses each person can run a day (25 unless you choose between 1 and 100) and each person's monthly fair-use limit.
  • Most active people this month, with Give more and Remove limit for each.
  • Need more AI?: what Advanced includes for your site, the Rainbook agent in Rovo on your Rovo credits, and a box for a top-up code from Rainbook support.

Only admins see this page. Nobody else ever sees a price, an upgrade offer or the word "budget".

Rainbook's AI is included in your subscription, with fair-use limits. Each person can run up to 25 Ask analyses a day, within a monthly fair-use limit for each person; questions that fit a report don't count. Summaries and other automatic AI draw on their own monthly pool. Your admins can lower the limits or give a person more. Reports, search and everything else keep working.

Automatic work (email and record summaries, AI-written health reasons, capture suggestions) has its own monthly pool, so a busy week of questions never stops your summaries. If that pool runs out, summaries show their last version with its date until the reset. If the site uses its month's included AI, Ask and drafts keep working within each person's daily analyses and monthly fair-use limit, up to a monthly ceiling for the site; past that, people's AI pauses until the reset. The Rainbook agent in Rovo keeps working throughout.

Rainbook's own AI, capture and background jobs are included in your subscription and use neither Rovo credits nor Automation steps. The Rainbook agent in Rovo, Rovo Studio agents that use Rainbook's tools, and Automation rules built from Rainbook's templates or actions use your Atlassian allowances: Rovo credits from 3 Dec 2026, and Automation steps from your first renewal on or after 3 Dec, shown per rule in your admin Usage tab.

Free on Standard for sites of 1–10 users: the whole CRM, industry packs, sidebars and Ask, with about 60 to 100 Ask analyses a month; background capture, summaries and drafts need a paid plan.

Admin controls

Settings › AI and Rovo holds every switch, and every change is recorded in the activity history.

  • Rainbook's own AI and Rainbook agent and tools in Rovo: two master switches, so you can turn either off without the other. When the agent is off, it replies that it is turned off for the site and to ask a Jira admin; Rovo still lists it.
  • One switch per feature: Ask, summaries, health reasons, drafts, the duplicate check, in-app proposals, and email summaries for AI clients.
  • Require a second approver and the thresholds for Home's insight cards. Personal limits are on Settings › AI budget.

For what each feature sends to the model and how long each thing is kept, read the AI Usage Policy. For everything else on data handling, see Security and privacy.

Teams in healthcare, financial services, government and legal can add the optional Regulated pack.