Atlassian Gives Rovo Chat Editable Memory, Agent Mentions and Shared Threads
Rovo users can inspect remembered context, call specialist agents into a conversation and hand complete chats to teammates, while administrators retain a site-level memory switch.
Edited by Tyronne Panaino
Atlassian introduced a set of Rovo Chat changes on September 17 that moves context from a hidden background feature into something users can inspect, hand off and reuse. The release adds visible memory controls, direct mentions for specialist agents, shareable conversations and custom skills for repeated workflows.
The practical change is continuity. A user can adjust what Rovo remembers, bring another agent into the current thread or pass the thread to a teammate without first rebuilding the work as a separate summary. Atlassian says access to a shared chat remains governed by the permissions the recipient already has.
Memory becomes an editable workspace control
Rovo previously used memory in the background to learn preferences and working patterns. The new memory view exposes that context so a user can inspect and edit what the assistant has retained, including response preferences and project details. Users can also adjust how Rovo responds.
The Atlassian announcement says memory can be switched on or off at both user and site level. That gives individuals a personal control while preserving an administrative boundary for organisations that do not want memory enabled across an instance. The announcement does not provide an independent privacy audit or a technical description of retention, deletion and isolation, so the control should not be read as proof of those wider properties.
Agents and teammates can inherit the thread
Rovo Chat can now call an agent with an @mention. Atlassian says the mentioned agent receives the thread, prior decisions and relevant artifacts, allowing a specialist agent to continue from the existing conversation rather than start with a fresh prompt.
The same continuity model applies to people. A chat can be shared with a teammate so the reasoning and context travel with the handoff. Atlassian states that the shared material is permissions-aware: a recipient should only see information they already have permission to access. That is an important product boundary, but the release note does not include independent testing of permission enforcement across connected applications. Administrators should therefore treat the documented behavior as the intended control and verify it against their own access model before relying on shared chats for sensitive work.
Custom skills turn repeated work into a reusable instruction set
The fourth addition lets teams describe a recurring process in natural language and turn it into a custom skill. Atlassian's example is a sprint-summary workflow built from a board, with Rovo asking follow-up questions about edge cases. Skills can be created, refined and updated in Rovo Chat or Rovo Studio.
Availability is not completely simultaneous. Atlassian says custom skills will reach all instances over the week following the September 17 announcement. The page presents the wider chat feature set as usable now, but does not publish a region-by-region or plan-by-plan matrix. A team that does not yet see skill creation should check its instance and administrative settings instead of assuming the rollout has failed.
What changes for teams
The release joins four related controls around context: what the system remembers, which agent can act on a thread, how a teammate receives the conversation and how a repeated process becomes a reusable skill. That can reduce manual context reconstruction, but Atlassian does not report measured time savings, recall accuracy or workflow-quality outcomes. Those results will depend on the connected data, permission design and the quality of each team's skill instructions.
Status
Confirmed. Atlassian announced the Rovo Chat features and the staged custom-skill availability. Confidence is medium because the evidence is a first-party product announcement without independent validation of behavior or outcomes.
Sources
Update note: Last reviewed 2026-09-22. We will revise this post if Atlassian changes availability, memory controls or the permission model for shared chats.
Sources
Drafted with AI assistance from source briefs; reviewed for citation completeness and label accuracy.