Atlassian Rovo

What it is

Rovo is Atlassian’s AI layer over the Teamwork Graph: chat, search, and buildable agents that read Jira, Confluence, Jira Service Management, Bitbucket and Loom natively, plus 75-plus third-party connectors including Google Drive, SharePoint, Slack, GitHub, Figma, Salesforce and Notion. It ships inside eligible paid Atlassian Cloud plans rather than as a separate purchase. Atlassian reported Rovo passing 5 million monthly active users in the quarter ended December 2025, and usage at more than 80% of the Fortune 500 by the quarter ended June 2026.

That bundling is why the page exists. The question every ops team asks — we already own Jira and Confluence, do we still need a separate knowledge layer? — now has a defensible “no” for a specific set of teams, and a defensible “yes” for the rest. The credit meter is what tells them apart.

Rovo is included, not free

Three access levels, and buyers conflate them:

• Rovo Free — what lands on your site automatically if you hold a paid Jira, Confluence or Jira Service Management plan. No separate purchase, no order form.
• Rovo Standard — standalone access for people with no paid Atlassian app seat: Rovo Chat and Search, Studio, browser extension, desktop and mobile apps, connectors, Slack and Microsoft Teams surfaces. In open beta and unbilled as of 5 September 2026, with a published rate of $5 per user per month per organization once beta ends. Orgs that do nothing revert to Rovo Free, and Atlassian gives at least 90 days to opt in.
• Rovo Dev — the coding agent in CLI, Bitbucket Cloud and GitHub, generally available and priced separately at $20 per developer per month with 2,000 Rovo Dev credits.

Included capacity is metered in Rovo credits, pooled at the organization level rather than capped per person:

Credits per user per month. A basic AI event — a Rovo Chat question, an agent run — costs 10 credits, and enriched cross-product context retrieval from the Teamwork Graph bills 1 to 10 credits per call on top.

Do that arithmetic before you plan a rollout. On Jira Standard, 25 credits is about two Rovo Chat questions per user per month. On Premium’s 70 it is about seven. Pooling saves the rollout — light users subsidise power users — but the allowance is sized for occasional use, not for “Rovo replaces how we find things.”

Overage is $0.01 per credit, or $10 per 1,000 credits, and extra usage is enabled by default. A team of 300 on Premium pools 21,000 credits a month; if real usage lands at 200 credits per person, the 39,000-credit overage costs about $390 a month, near $4.7K a year. Against Glean at roughly $50 per seat with a 100-seat floor — about $180K a year for those same 300 people — the two are not in the same order of magnitude. Price is not the deciding variable here. Coverage and answer quality are.

Best for

An ops or IT leader at a company where Jira and Confluence are already the system of record, on Premium or Enterprise plans, whose retrieval problem is mostly “find the decision, the ticket and the runbook” rather than “search every SaaS the company owns.” The scoped use case where Rovo is clearly the right call: service-desk deflection on Jira Service Management, where the Service Collection allowance of 250 to 1,500 credits per user is sized for actual agent volume and the content being searched already lives in Atlassian.

Not for

• Teams whose knowledge lives outside Atlassian. If the answers are in Google Drive, Slack threads and Salesforce records and Confluence is a graveyard, Rovo inherits that. Retrieval quality is a function of what you wrote down.
• Companies without paid Atlassian Cloud plans. Rovo Standard at $5 works, but you are buying a knowledge tool from a vendor whose gravity is elsewhere.
• Server and Data Center holdouts. Rovo is Cloud-only, and Atlassian’s Cloud-only direction means the migration is the project, not the AI.
• Anyone who needs connector breadth as the binding constraint. Glean’s catalog is deeper, and 75-plus connectors that miss your system of record is the same as zero.

Versus the alternatives

• Glean — the category share leader and the head-to-head in every 2026 alternatives guide. Pick Glean when retrieval must span dozens of systems with none privileged, permissions enforcement is already clean, and the budget clears a roughly $60K first-year floor. Pick Rovo when Atlassian holds most of the corpus and you would be paying Glean to index what you already own.
• Microsoft 365 Copilot — the other incumbent by installed base, at $30 per user per month. The rule is symmetric to Rovo’s: if the work product is documents, email and Teams threads, Copilot’s home-field advantage beats Rovo’s. If the work product is tickets, epics and specs, Rovo’s does. Companies running both end up buying both and should decide which one owns “ask a question” before either rolls out.
• Dust — the fast-growing AI-native entrant, agent-first rather than index-first. Pick Dust when the goal is shipping assistants to individual teams on a short cycle; pick Rovo when the graph, not the assistant, is the asset. Dust vs Glean works through that axis.

If none of these fit, the problem is the corpus, not the search over it — best knowledge bases for teams covers the case where the content was never written, and RAG explains what retrieval can and cannot recover.

Watch-outs

• The credit meter is the real price, and the default setting bills you for it. “Extra usage” is on out of the box, so overage accrues silently at $0.01 per credit and arrives at month end. Guard: before the rollout, set a usage budget in the admin console and decide deliberately whether to leave extra usage on. Then measure a two-week pilot cohort’s credits per user and multiply by headcount before announcing Rovo company-wide — the pilot number, not the allowance, is your forecast.
• Rovo Standard’s $5 is a post-beta price, not a locked one. The beta is unbilled, and a plan priced during a land phase is the plan most exposed to repricing. Guard: if Rovo Standard is load-bearing in your budget, write the reversion path into the decision doc now — Free is where you land by default — and set a calendar check on the 90-day opt-in window rather than discovering it in a renewal email.
• Prompt injection against Rovo has been demonstrated twice in 2026. Researchers disclosed a one-click URL-parameter attack using a crafted rovoChatPrompt, which Atlassian fixed server-side on 8 July 2026, and a separate content-borne injection where hidden instructions inside a document cause Rovo to append search results to an attacker-controlled URL — reported unresolved as of 5 August 2026. There is no published evidence of abuse against real organizations. Guard: treat any Jira or Confluence surface that accepts external input — a public service desk portal, a customer-facing form — as untrusted content that Rovo will read, and scope connectors so the agent’s reachable data is smaller than the user’s. Ask Atlassian for the current status of the content-borne vector in writing during procurement.
• Answer quality tracks Confluence hygiene, and nobody audits that first. Rovo will confidently surface a three-year-old page because nothing marked it stale. Guard: run ten questions with known-correct answers against the live instance during evaluation and check each cited source’s last-edited date. If half the citations are stale, the project is a content-cleanup project wearing an AI budget.
• Rovo Dev is a separate line item, and org charts hide it. Teams read “Rovo is included” and plan for the coding agent too. Guard: budget Rovo Dev at $20 per developer per month as its own row, with its own 2,000-credit pool that does not share the Rovo meter.

Related: the enterprise AI rollout stack prices the layers that sit around a retrieval tool like this one.