Atlassian Code Context Security: Permissions, Indexing, and Data
On this page
Quick answer
Atlassian Code Context is off by default and requires organization-admin opt-in. It indexes all source files across all branches and tags in selected GitHub Cloud or Bitbucket Cloud repositories, stores file metadata, code chunks, and vector embeddings in an Atlassian-managed isolated environment in the United States, and validates SCM permissions at query time.
Those controls do not make the feature a security scanner or eliminate source-governance work. Before enabling, remove committed secrets, confirm indexing rights and US data handling, correct SCM permissions, approve third-party agent providers, and test stale results, revocation, disablement, and deletion.
What is stored
| Data | Officially documented boundary | Control question |
|---|---|---|
| Source files | All files across all branches and tags in opted-in repositories | Are every branch, tag, generated file, and historical secret authorized? |
| File metadata | Paths, names, repository IDs, branch and tag references | Can names or structure expose sensitive architecture? |
| Code chunks | Files split for retrieval | Who may query and receive each chunk? |
| Vector embeddings | Numerical representations created by a code-optimized model | Are storage region, retention, deletion, and model processing approved? |
| Third-party context | Retrieved source sent through TWG CLI to selected agents | Are provider terms, logs, training, retention, region, costs, and actions approved? |
Source: Atlassian Support: what is Code Context, reviewed August 27, 2026.
Admin opt-in and repository rights
Only an organization admin can enable Code Context for an eligible site. Atlassian says individual contributors cannot exclude their code after a repository is enabled. That makes pre-enable scope review essential.
Confirm the organization has rights to mirror, index, store, chunk, and embed the code. Include vendor and open-source license obligations, customer code, generated code, forks, archived repositories, branch history, and tags. Do not assume that repository read access alone grants permission for AI processing.
Use the GitHub vs. Bitbucket setup guide for connection-specific checks.
Permission enforcement
Atlassian says results are restricted to repositories and branches the querying user can access, with validation against the SCM permission endpoint at query time. Test the claim within your configuration:
- Users with full, partial, and no repository access.
- Branch-level restrictions and inherited groups.
- Access removed during an active session.
- Transferred, archived, renamed, or newly private repositories.
- Connector or SCM permission-endpoint outages.
- Service accounts and authorized agents.
Permission-aware retrieval prevents some excess access; it does not ensure the user is allowed to send retrieved code to a third-party model or use it for a new purpose.
Indexing, freshness, and relevance
Initial indexing can take up to 24 hours, and Atlassian currently provides no repository-level indexing indicator. Pushes and branch updates trigger incremental indexing, but recently changed code may not appear immediately.
Semantic search ranks conceptual similarity and has no current confidence signal. It may underperform on niche languages, non-English comments, or ambiguous questions, and comments can influence relevance. Require citations to repository, path, symbol, branch, and revision, then verify the source before acting.
Never use semantic search as the only vulnerability or compliance control. Atlassian explicitly says it is not a replacement for dedicated scanning.
Data location and eligibility limits
Atlassian says stored Code Context data is in the United States and that data-residency configuration is not currently supported. The setup guide tells organizations with non-US residency or sovereignty requirements to confirm whether enabling is appropriate.
The current limitations also say Code Context is unavailable for HIPAA-enabled sites and is not currently HIPAA compliant. Do not infer eligibility from another Rovo feature or Atlassian plan.
Third-party agent transfer
When TWG CLI augments Codex, Claude Code, or another third-party agent, source code is sent to that provider. The provider boundary may include its own identity, model, region, retention, training policy, logs, subprocessors, costs, API limits, and action tools.
Approve retrieval and action authority separately. A provider may receive a permitted snippet while still being forbidden to write files, invoke tools, contact a network destination, or modify another repository.
See the TWG CLI coding-agent guide.
Disablement and deletion
Atlassian says disabling Code Context stops indexing immediately and deletes stored embeddings, indices, and code chunks. Test who can disable it, what happens to queries during the transition, how deletion is evidenced, and which copies remain outside Code Context.
Third-party prompts, outputs, logs, local caches, benchmark reports, exported snippets, or pull requests may require separate retention and deletion procedures.
Readiness gates
- Secrets scan completed and credentials rotated.
- Repository, branch, tag, fork, and archive scope approved.
- Indexing and embedding rights documented.
- US storage and processing approved.
- SCM permissions reviewed and denial-tested.
- Third-party providers and models approved separately.
- Freshness and relevance failures tested.
- Human review and repository protections preserved.
- Disablement, connector revocation, user removal, and deletion tested.
Run the Atlassian Code Context readiness check to record the pilot gate locally.
Frequently asked questions
Is source code used to train Atlassian AI models?
Atlassian says Code Context source is not used to train Atlassian’s AI models. Third-party providers have separate terms that must be reviewed.
Are embeddings reversible into source code?
Atlassian describes embeddings as non-human-readable and not intended to be reversible. They remain stored derived data that requires approved processing, access, retention, and deletion.
Does disabling delete every copy everywhere?
Atlassian says its Code Context embeddings, indices, and code chunks are deleted. Separately manage content sent to third-party agents, logs, exports, local caches, and downstream changes.
Does query-time permission validation guarantee data governance?
No. It enforces an SCM access boundary, not purpose limitation, third-party transfer approval, output correctness, or downstream action authority.
Official sources
- Atlassian Support: what is Code Context
- Atlassian Support: set up Code Context
- Atlassian: introducing Code Context
Source check: August 27, 2026. Recheck indexing scope, storage region, permissions, provider transfer, limitations, lifecycle, and deletion behavior.