30. 12. 2025 Fabrizio Dovesi Atlassian, Service Management

Atlassian Rovo Today: Architecture, Technologies, and Enterprise Trust

The current architecture and underlying technologies behind the Atlassian Rovo engine, in light of recent developments

The last few months showing the path forward

If you have been following the evolution of AI in recent years, you know changes come fast and often, and even a single year can be too long an interval to re evaluate topics already analyzed. In the space of a month, new frontiers fall, new challenges are solved, and fresh ones emerge. For this reason, in this article we focus on the progress Atlassian Rovo has made in the final stretch of the year. Over the past four months, Rovo has undergone a series of upgrades that clearly signal its strategic direction. Search relevance has improved significantly through advanced re-ranking techniques, personalization, and space-aware filtering, making information discovery faster and more accurate across tools. These changes build on a more mature retrieval architecture, moving beyond basic AI search toward learning-driven, context-aware results. Rovo has evolved into a full stack AI platform that blends cross suite semantic search, context aware chat, and actionable agents across Atlassian and connected tools.

In October 2025, the “Rovo everywhere” initiative has expanded the platform’s reach with Canvas, personal memory, and stronger governance features, enabling safer and more controlled agentic workflows. The data ecosystem around Rovo has continued to grow through new connectors and enterprise synchronization, enriching the context available to the Teamwork Graph.
Finally, in December 2025, Atlassian also acquired Secoda to strengthen semantic discovery over structured data for Rovo, further aligning the platform with end to end insights over enterprise data sets. The acquisition of Secoda highlights a strategic focus on structured and hybrid data, positioning Rovo as a broader, enterprise-grade intelligence layer rather than a standalone assistant.

In the following sections, we analyze the current architecture and the underlying technologies behind the Rovo engine, taking into account the recent innovations introduced over the past few months that have shaped its present structure.

The data layer, Teamwork Graph and connectors

At the core of Rovo sits the Teamwork Graph, a graph based intelligence layer mapping people, teams, goals, issues, pages, files, events, and signals across Atlassian and third party SaaS. This layer provides the organizational context that powers Search, Chat, and Agents, so answers and actions are relevant to the user, their permissions, and their work. Atlassian documents the Teamwork Graph as the foundation for Rovo’s personalized knowledge experiences and for connecting external sources at scale (more details on Teamwork Graph)

Rovo ingests data through multiple connector types, each with distinct behavior and trade offs.

Synced and Synced Lite connectors index content and metadata into the graph,
Direct connectors retrieve results live via external search APIs without indexing into Atlassian,
Smart Link connectors render URL previews and extract searchable signals with no admin setup.

This design impacts latency, freshness, and depth of context available to Search, Chat, and Agents, while preserving source application permissions. Atlassian’s admin guide and support pages describe connector setup, the permission model, and recommended sources to connect first, for example Google Drive, SharePoint, Slack, Notion, and others (see how to manage Rovo connectors here).

Recent updates added automatic discovery from more Smart Link sources, plus searchable calendars from Google and Outlook, which improves cross tool visibility without heavy configuration overhead. Beyond native connectors, Unito announced enterprise grade connectors (more details here) that synchronize Salesforce, ServiceNow, Asana, and Wrike into the Teamwork Graph, enriching Rovo with cross system context for Search, Chat, and Agents. Atlassian’s own blog highlights how these connectors supercharge Rovo’s intelligence across the organization. Rovo also ships a native app for Slack, which supports DMs, mentions, right rail assistance, and channel triggers, while the Slack connector makes chat content discoverable in Rovo with source permissions enforced.

Search architecture, indexing, OpenSearch, BM25 plus kNN, and re ranking

Rovo implements a hybrid retrieval stack that combines keyword based ranking, BM25 (Best-Matching 25), with semantic search via embeddings and kNN (k Nearest Neighbors), serving queries on AWS OpenSearch. This architecture underpins full page search, smart answers with sources, and context fetching for Chat and Agents. Atlassian Engineering describes how the relevance stack works, what user signals are used, and how content is modeled and indexed for fast discovery across connected apps. In September 2025 the ranking model was upgraded with pairwise loss training, and Rovo introduced personalized prompt suggestions and space filters that help users narrow results to their most used spaces. These changes aim to reduce noise and improve precision in multi tool environments, and they are documented in partner summaries of the monthly updates. From a user perspective, Rovo delivers smart answers with citations, knowledge cards for people and definitions, advanced filters, and a browser extension that brings the Rovo Search bar to every new tab, which reduces context switching in daily activities.

AI orchestration and Rovo Agents, multi LLM, Studio, and Forge

Rovo orchestrates multiple LLMs, combining open source and self hosted models with third party providers such as OpenAI, Anthropic, Google, and Mistral or Llama. For enterprise data protection, Atlassian states that these providers do not store or use customer inputs or outputs to train their services, and that AI features inherit Atlassian Cloud Platform security and privacy practices. The Trust Center and the Rovo data usage and privacy page describe these commitments, data flows, and controls. Rovo Agents are configurable AI teammates that can be invoked in Chat, automation rules, or during editing in Jira, Confluence, and Jira Service Management. They read from connected sources while respecting permissions, and they can perform actions such as creating or updating issues, drafting release notes, triaging service requests, or composing documents. Atlassian ships out of the box agents and provides a builder experience through Studio for custom agents and workflows (check the How to use Rovo guide for further details). As mentioned before, in October 2025 Atlassian added Canvas, a collaborative surface inside Chat, personal memory for the assistant, and deeper governance in Studio, including fine grained permissions, conversation logs, analytics, and testing before deployment. Atlassian also announced upcoming MCP support to bring third party capabilities, for example GitHub, Figma, Box, HubSpot, directly into agent workflows. Developers can extend Rovo via Forge using the rovo agent module and custom actions, with serverless execution, secure permissions, and distribution via Marketplace or private channels. Atlassian provides a “hello world” tutorial and guidance for building agents that invoke custom logic and integrate enterprise data safely.

Security, governance, and data residency

Enterprise adoption requires robust trust controls. Atlassian highlights SOC 2 and ISO 27001 compliance, restrictive data use policies for AI features, and centralized administration. For Rovo, audit logs capture admin and user actions, for example chat started, agent created or updated, connector changes, and these events are integrated with Atlassian Guard. The Trust Center and Rovo privacy guide outline how models are selected, how inputs and outputs are processed, and how access is enforced end to end. On data residency, Atlassian introduced Rovo residency in limited availability during 2025, aligning with existing controls in Jira, Confluence, and Jira Service Management. The community announcement and core residency documentation explain pinning to regions, scheduling moves, and the operational considerations for large datasets. Permission models from source systems are honored throughout, so users only see what they are allowed to access across Atlassian and connected apps (more details on Rovo residency announcement here). Note, in August 2025 Anthropic updated its policy for consumer chats, opt out required to avoid training use, while maintaining distinct enterprise or API assurances. This underscores the importance of enterprise contracts and provider routes used by Atlassian for Rovo (for further details, see the AI2 Work analysis).

Conclusion

In conclusion, as AI evolves at a month by month pace, this review should help clarify Atlassian Rovo’s latest end of year advances and set a baseline for what comes next.

Rovo now operates as an AI system of work, combining stronger search, context aware chat, and actionable agents on the Teamwork Graph for faster, more accurate decisions. “Rovo everywhere” with Canvas, personal memory, and Studio governance improves control and rollout, while MCP and the Secoda acquisition expand third party and structured data reach. Next steps: connect high value sources, enforce Studio governance, and start with agents that reduce noise and speed handoffs. For 2026 planning, Rovo provides a concrete blueprint, connect context, models, and actions inside a single, governable platform.

Ultimately, it is prudent to keep a close eye on this fast moving technology race among leading players, Atlassian’s Rovo included.

These Solutions are Engineered by Humans

Did you find this article interesting? Does it match your skill set? Our customers often present us with problems that need customized solutions. In fact, we’re currently hiring for roles just like this and others here at Würth IT Italy.