Atlassian

Rovo agents act in real systems. They can "use plugins and access data from other Atlassian and connected third-party apps" and "perform specialized skills like organize, create, or edit Jira work items or Confluence pages." Admins "connect Atlassian to tools like Figma, Box, HubSpot, and more" in Admin Hub. The docs carry third party MCP agents that work out of the box, with usage guidelines, and partners' Marketplace agents can be installed. Rovo connects to 50+ applications, including Google Drive, Slack, GitHub, Salesforce, Notion, Teams and SharePoint.

These are actions, not only reads: agents create and edit work items and pages through native integrations, connected third party apps and MCP. Credential scoping and rotation are covered only by the governance page.

Sourcesupport.atlassian.com/rovo/docs/agents; readread 2026-09-17

Customers compose agents and run them inside Atlassian's rule engine. They "create and edit Rovo agents" through chat or manual setup, writing instructions and adding subagents with their own knowledge sources, along with tools and skills, and they can set a reasoning tier and duplicate and share agents.

The docs describe delegating specialized work to subagents and running agents in automation rules and in the Studio app. Agents invoked from Jira and Confluence automation rules put agent steps inside the platform's rule engine, and delegating to subagents allows reuse and composition. Agents in Jira can take ownership of work items, and an Incident Command Center brings detection, investigation and resolution together.

Retry and fallback behavior and agent versioning are not documented.

Sourcesupport.atlassian.com/rovo/docs/agents; readread 2026-09-17

Agents are grounded in a maintained graph over the customer's content, with sources set per agent. Knowledge sources for Rovo subagents let customers configure agents with "custom knowledge sources like Confluence, Jira, and Google Drive, and by enabling deep research or web search."

Agents reach data in other Atlassian and connected third party apps depending on their knowledge sources, Rovo Chat documents supported data sources and a Deep research mode, and memory management describes a summary built from "Teamwork Graph activity across Atlassian products." The Teamwork Graph is one layer over work and people, decisions and code.

Sources are connected rather than trained into the model, and the graph persists across Atlassian and third party content. How sources refresh and how answers cite them are not documented.

Sourcesupport.atlassian.com/rovo/docs/agents; readread 2026-09-17

Rovo acts outside chat only with the user's confirmation. The agents documentation says agents "perform specialized skills (with your permission) like organize, create, or edit Jira work items or Confluence pages," and both the tools and skills pages say they "allow them to do things outside of chat with your confirmation." Rovo Service offers autonomous or supervised level one support as a configured choice, which sets autonomy by risk.

Whether confirmation can be waived per tool, and whether approvals can be routed to other channels, is not documented. Studio admins also control who can create agents: all users, up to ten selected groups, or admins only.

Sourcesupport.atlassian.com/rovo/docs/agents, /docs/rovo-agent-permissions-and-governance; readread 2026-09-17

Atlassian documents a third party attestation and an access surface. The Compliance Resource Center entry for FedRAMP, typed Certifications / Attestations, states "Jira, Confluence, and Jira Service Management have achieved FedRAMP certification (formerly known as FedRAMP Moderate authorization, now Certification Class C)," an authorization that requires assessment by an accredited third party assessment organization.

For access, the trust pages describe identity and access management across Atlassian products for the whole organization, with Atlassian Guard used to "connect an identity provider, enforce SSO and MFA, manage external user security," on a stated zero trust approach. Rovo agent governance adds Studio admin control of who may create agents (all users, up to ten selected groups, or admins only).

The Compliance Resource Center carries a fuller certification list, with SOC and ISO entries. The FedRAMP certification covers Jira, Confluence and JSM, and Rovo Studio is "not available" on Atlassian Government Cloud.

Sourceatlassian.com/trust/compliance/resources/fedramp, /trust/data-protection; support.atlassian.com/rovo/docs/rovo-agent-permissions-and-governance; readread 2026-09-17

Teams can inspect agent runs after the fact. The Rovo agents docs cover "track how often a Rovo agent is used," "debug responses from Rovo agents" and "review live conversations with your Rovo agent," and Rovo Dev documents viewing credit usage and session management. Agents in Jira can be assigned work items "with full audit logging," so each agent action lands in the same tracker history people use, which keeps the audit record apart from runtime traces.

Whether debugging and review list tool calls one by one is not documented, and neither is export to monitoring systems.

Sourcesupport.atlassian.com/rovo/docs/agents; atlassian.com press release (read 8 August 2026); readread 2026-09-17

Rovo has a documented memory layer with a stated scope. "Rovo memory management gives you visibility and control over what Rovo knows about you," built on transparency ("you can always see what Rovo knows about you") and control ("you decide what instructions get saved or removed").

There are two types. Implicit memory is a summary built from Teamwork Graph activity, which the user can suggest edits to. Explicit memory is specific instructions captured in chat ("Remember that I prefer direct answers"), which the user "can view, edit, or delete at any time." The scope is stated: "Each memory is scoped to you and to the specific Atlassian site you're using; memories on one site are separate from memories on another."

Details in the implicit summary cannot be removed directly, only edited by suggestion, and memory is kept per user, not per team.

Sourcesupport.atlassian.com/rovo/docs/what-is-rovo-memory-management; readread 2026-09-17

Rovo runs only in the cloud. Agent governance lists availability as "Atlassian Cloud: Standard, Premium, Enterprise" and "Atlassian Government Cloud: not available," so the agent layer is absent from the government environment and from Data Center deployments. The trust pages state products are "hosted on AWS," which says where data sits but offers no choice.

No region choice for Rovo data is documented here, and the data residency page under /trust returns a 404. Atlassian's data residency documentation on pinning data to a realm may say more.

Sourcesupport.atlassian.com/rovo/docs/rovo-agent-permissions-and-governance; atlassian.com/trust/data-protection; readread 2026-09-17

Atlassian ships a browsable agent directory, named agents and an installable partner catalog.

The Rovo agents docs carry "Browse Rovo agents" and "Out-of-the-box agents" pages, guides for named agents (the Ops Expert, the Service Request Helper, the Request Router and the Jira Delivery agent), third party MCP agents that work out of the box, and "Marketplace agents built by third-party Marketplace partners available to be installed."

Agents can be duplicated and shared across the organization, and each named agent stands on its own. Customers edit them by duplicating and changing instructions, tools and knowledge sources.

No agent count is given.

Sourcesupport.atlassian.com/rovo/docs/agents (documentation index); readread 2026-09-17

Work reaches Rovo agents without anyone asking. Agents "can be accessed in chat, in automation rules, while editing (via the shortcut /rovo or /ai) in Confluence and Jira, or in the Studio app," and the docs carry an "Automate Rovo agents" page plus a Request Router agent for incoming requests. Agents engage when assigned a work item, when mentioned in a comment, or are "automatically tapped in when work moves into a designated status." An automation rule firing on a status change reaches the agent with nobody starting it, alongside scheduled automation rules and conversation.

The Automate Rovo agents page may list further triggers.

Sourcesupport.atlassian.com/rovo/docs/agents; atlassian.com/blog Team 26 (read 8 August 2026); readread 2026-09-17

Model choice exists in one developer surface. "Rovo Dev CLI offers several different large language models (LLMs) so you can choose the right one for the work you're doing," switched with the /models command, each carrying a credit multiplier against the default model, with guidance to pick simpler models for documentation and advanced ones for complex code generation. Elsewhere Rovo offers a reasoning tier per agent and reasoning modes in Chat, which set effort, not the model.

The choice exists only in the Rovo Dev CLI. No model or provider selection is documented for Rovo Chat or for Rovo agents in Jira, Confluence or JSM, the models themselves are not named, and no bring your own model or provider account is offered.

Sourcesupport.atlassian.com/rovo/docs/switch-between-large-language-models-in-rovo-dev-cli, /docs/agents; readread 2026-09-17

Rovo both consumes MCP servers and runs as a server other tooling drives. The Rovo Dev documentation covers "Rovo Dev and Model Context Protocol (MCP)," "connect to an MCP server in Rovo Dev CLI," "use server mode in Rovo Dev CLI" and "extend Rovo Dev CLI with agent skills." Rovo also hosts third party MCP agents that work out of the box and partners' Marketplace agents inside Atlassian products, and customers build their own agents with tools and skills. Atlassian opened the Teamwork Graph to third parties at Team 26.

So MCP runs in both directions, customers add their own tools, and Rovo fits into other agent systems, since third party agents work inside Atlassian products. The Atlassian REST API and the Forge developer platform sit outside these docs.

Sourcesupport.atlassian.com/rovo/docs/agents (documentation index); readread 2026-09-17

Rovo gives customers a documented way to evaluate agents. The documentation set includes "Evaluate the performance of your Rovo agent," "Debug responses from Rovo agents" and "Review live conversations with your Rovo agent," alongside usage tracking and a process for verifying agents ("What's a verified Rovo agent," "Verify Rovo agents in your organization"). Evaluating performance, debugging responses and reviewing live conversations report on agent output, and verification adds a quality gate before an agent is promoted across the organization.

Scoring mechanics and testing against fixtures or datasets are not documented.

Sourcesupport.atlassian.com/rovo/docs/agents (documentation index); readread 2026-09-17

Rovo agents work through product APIs and connectors, not screens. They act in Chat and automation rules, in the Confluence and Jira editors and the Studio app, and through plugins and connected third party apps. The documentation index lists no browser, desktop or computer use tool, and the nearest surface, Rovo Dev CLI, runs in a terminal and in the IDE; terminal and shell access are not computer use. A change to a user interface has nothing to break here.

Sourcesupport.atlassian.com/rovo/docs/agents; readread 2026-09-17