Atlassian has introduced Code Context, an open-beta feature designed to give AI coding agents access to code repositories alongside the broader organizational information stored in its Teamwork Graph.
The company says the feature can help AI agents work with a fuller picture of a software project by connecting source code with information from tools such as Jira, Confluence and other connected services.
Code Context is currently available in open beta for GitHub users in Rovo Chat and through Atlassian’s Teamwork Graph CLI, with Bitbucket support rolling out. Atlassian says the current experience could change before general availability.
The feature is intended to address a limitation in AI-assisted software development: an agent may be able to inspect the repository available locally but lack information about related repositories, project requirements, documentation or previous technical decisions.
With Code Context, Atlassian builds a searchable index from connected repositories after an organization administrator explicitly enables the feature.
The system divides files into smaller sections and converts those sections into vector embeddings. The code chunks and embeddings are then stored in a secure, isolated Atlassian environment.
Repository pushes and branch updates trigger incremental re-indexing of changed files. Atlassian cautions that indexing is frequent rather than instantaneous, so recently pushed code may not immediately appear in search results.
Code Context combines lexical search with semantic search. Lexical search is intended for exact terms, such as a particular function name or string, while semantic search is designed to find code based on the meaning of a request rather than an exact textual match.
Atlassian says the underlying agent can automatically select the more appropriate search method and can combine both approaches when needed.
The company has positioned the feature as more than a repository search system. The Teamwork Graph connects information across Atlassian products and external tools, including Jira, Confluence, GitHub, Google Drive and Slack, giving agents access to relationships between people, content, activities and work.
Through the Teamwork Graph CLI, Atlassian says AI agents including Claude Code, Codex, Gemini and Cursor can query Atlassian data and Teamwork Graph information.
Atlassian’s documentation describes the Teamwork Graph as a unified data layer connecting teamwork information across Atlassian applications and external tools.
The company says this broader context can help an agent investigate problems that extend beyond the repository a developer currently has open.
For example, an issue in one service could be connected to another repository, a Jira work item or documentation that explains an earlier architectural decision. Code Context is intended to make those relationships available during an AI-assisted coding task.
Atlassian reported that its internal testing found agents using additional Teamwork Graph context produced results that were 44% more accurate while using 48% fewer tokens than agents without that additional context.
Those figures are Atlassian’s own internal results. They should therefore be viewed as company-reported performance figures rather than an independently verified benchmark.
Atlassian says the reduction in token use can come from giving an agent more relevant information upfront, reducing the amount of searching and context gathering required before it can work on a task.
Security and access controls are central to the feature. Code Context is disabled by default and must be enabled by an organization administrator.
According to Atlassian’s Code Context documentation, users only receive results from repositories and branches that they already have permission to access in their source-control system.
Atlassian says it validates those permissions at query time against the repository’s permission endpoint, meaning Code Context does not provide users with access beyond their existing source-control permissions.
There is no individual contributor opt-out once an organization administrator enables indexing for a repository. Atlassian says this is consistent with how its existing Bitbucket lexical search works.
For GitHub repositories, Atlassian says code is mirrored into Atlassian infrastructure before indexing. For Bitbucket repositories, the company says code is indexed directly.
The indexed information includes source-code files, file metadata, code chunks and vector embeddings. Atlassian says that for opted-in GitHub Cloud and Bitbucket Cloud repositories, all files across all branches and tags are included.
The indexed code and embeddings are currently stored and processed in the United States.
Atlassian says data residency configuration is not currently supported for Code Context. Organizations with data-residency or data-sovereignty requirements outside the United States are advised to assess whether the feature is appropriate for their environment.
The company also recommends scanning repositories for exposed secrets before enabling Code Context. Hardcoded API keys, passwords, tokens and other sensitive information contained in opted-in repositories can be indexed and may be discoverable by users who already have access to those repositories.
Atlassian further advises organizations to confirm that they have the necessary rights and approvals to allow the company to mirror, index and store their source code.
Atlassian says code indexed through Code Context is not used to train its AI models.
The company does use a code-optimized embedding model to create vector representations of source code. When Code Context is used through Rovo Chat, relevant code snippets may also be provided to a large language model as context for generating an answer.
There is an additional consideration when developers use third-party coding agents.
Atlassian says that when Code Context is accessed through the Teamwork Graph CLI with products such as Claude Code or Codex, source code can be sent to those third-party products and will be subject to their respective terms.
Atlassian places responsibility on organizations to evaluate the security and suitability of those third-party providers, as well as their costs, API capacity and agent actions.
The company also warns that AI-generated responses remain probabilistic. It recommends that users verify AI-generated answers before acting on them, particularly during critical activities such as incident response and security decisions.
Atlassian says semantic search is intended as a discovery tool and should not replace dedicated security or compliance tools used for auditing and vulnerability detection.
The feature has several current limitations. Atlassian says semantic search does not provide exact-match results, recently pushed code may be delayed, and performance is strongest with widely used languages such as Python, Java and TypeScript.
The company says performance can be weaker with niche programming languages, non-English code comments and highly ambiguous queries. Code Context also does not currently provide a confidence signal showing how certain the system is that a result is relevant.
New code may also stop being indexed during an outage affecting a connected source-control system until the connection is restored.
At present, GitHub Cloud and Bitbucket Cloud are the supported source-control systems, while Atlassian says GitLab and other systems are planned for future releases. Code Context is also unavailable to HIPAA-enabled sites.
Atlassian’s Teamwork Graph CLI provides the broader interface through which AI agents can interact with Atlassian’s data and services. The company says the CLI can work with Claude Code, Codex, Gemini, Cursor and other supported agents.
Atlassian’s developer documentation says the CLI installs agent skills that help coding agents work with Teamwork Graph and Atlassian products. Codex, Cursor and Gemini can use the universal .agents/skills location, while Claude Code can receive the skills under .claude/skills.
The Teamwork Graph itself connects information from products including Jira, Confluence, Bitbucket, Jira Service Management, Loom, Trello and Rovo, while also supporting connections to external services.
Atlassian is effectively positioning this infrastructure as a context layer that can be accessed by different AI agents rather than tying developers to a single AI coding product.
The company currently provides the Teamwork Graph CLI during beta without a separate CLI charge, although Atlassian’s command documentation identifies certain enriched capabilities that require Rovo Credits.
Atlassian has said it will provide at least 90 days’ notice before introducing charges for currently free CLI usage where applicable.
Discover more from Aree Blog
Subscribe now to keep reading and get access to the full archive.
