Atlassian's Customer Agreement, effective starting October 1, 2026, states that the customer owns all intellectual property and other rights in Customer Data, and Atlassian's data contribution page, effective August 17, 2026, names Jira, Confluence and Jira Service Management in its initial scope and does not name Bitbucket (read November 17, 2026).
Bitbucket is one source in AI training data licensing, and two programs name it among the code hosts they accept. Not licensing is a valid result.
The DataPacts Editorial Team dates every term and program statement below; the method is in how we source, date and check every figure.
What a Bitbucket Workspace Holds and Why Buyers Name It
A Bitbucket workspace holds projects and repositories with their commit history, branches, pull requests and review comments, a wiki per repository and Pipelines definitions, and two programs name Bitbucket among the code hosts they accept.
Mercor lists Bitbucket among the tools named in its Software Company listing, and Pangea.ai, a marketplace, states that owners connect GitHub, GitLab or Bitbucket read-only (both checked November 9, 2026). Atlassian Corp (SEC CIK 0001650372) owns the platform.
Code as a data type, with acquirers, rates and eligibility, is covered in selling or licensing source code and codebases; this page stays on what is specific to Bitbucket.
Who Controls Bitbucket Data in a License
Four layers decide what Bitbucket data a company can license: its agreement with Atlassian, Atlassian's AI commitments and data contribution settings, the licenses and authors inside the code, and the people and secrets in the history.
The table compares the four layers (as of 2026-11-17).
| Layer | Who | What the record shows | Effect on a license | Source |
|---|---|---|---|---|
| Customer agreement | the licensing company and Atlassian | Customer Agreement effective starting 2026-10-01: customer owns rights in Customer Data; retrieval and deletion follow the Documentation | the company controls its repositories; Atlassian's processing rights and exclusivity are a question for counsel | atlassian.com/legal/atlassian-customer-agreement |
| Bitbucket Cloud terms and AI | Atlassian | Product-Specific Terms s. 3; AI Terms effective 2026-08-17; data contribution scope does not name Bitbucket | covers Atlassian's own use and the repository rules, not a buyer's | atlassian.com/legal/product-terms; atlassian.com/legal/ai-terms; atlassian.com/trust/ai/data-contribution |
| Code and authors | employees, contractors, outside code | licenses of public repositories and third-party code (Product-Specific Terms s. 3.1, 3.3) | ownership and license checks before delivery | code-ownership pages |
| People and secrets in history | staff, customers, credential owners | names, emails, committed secrets | removal before delivery | de-identification pages |
Customer Agreement: Ownership, Retrieval and Deletion
The Atlassian Customer Agreement, effective starting October 1, 2026, states in section 13 that as between the parties the customer owns all intellectual property and other rights in Customer Data.
Section 4.4 says the Documentation describes how the customer may retrieve its Customer Data from the Cloud Products, and section 12.4 says that after expiration or termination Atlassian deletes Customer Data in accordance with the Documentation, with no period stated in the clause.
Section 6.1 makes the customer responsible for the rights and consents Atlassian needs to use Customer Data (read November 17, 2026). Whether those processing rights conflict with a buyer's no-prior-grant term is a question for counsel and the buyer, and the same holds for exclusive vs non-exclusive data licenses.
Product-Specific Terms, AI Terms and Data Contribution
Atlassian's Product-Specific Terms add four Bitbucket Cloud rules in section 3, and its AI Terms and data contribution page, both effective August 17, 2026, do not name Bitbucket repositories as contributed data.
The four rules are that each repository is designated public or private with a license stated for each public repository (3.1), that Atlassian may remove content consuming unreasonable storage unrelated to a source code repository's purpose, which means removing the entire repository (3.2), that the customer reads the licenses of third-party code it accesses (3.3), and that permissions granted to users cannot be prevented by Atlassian and some actions are irreversible (3.4).
AI Terms section 2.1 treats Input and Output as Customer Data and bars Atlassian's subcontractors from using them to train or improve their models. Bitbucket is not named in the data contribution scope read, and these terms cover what Atlassian does, not what a buyer may do with an export.
How Repositories Leave Bitbucket
Bitbucket repositories reach a buyer by three routes: a Git clone the company runs, the REST API for what Git does not carry, or a read-only connection a program sets up.
The table compares the routes (as of 2026-11-17).
| Route | Who may run it | What it carries | Limits | Source |
|---|---|---|---|---|
| Git clone over HTTPS or SSH | a user with repository access and an SSH key or Git Credential Manager | files and Git metadata: commits, branches, tags | repository cap 4.0 GB, recommended under 2.0 GB; 60,000 Git requests per hour | support.atlassian.com/bitbucket-cloud/docs/clone-a-git-repository/ |
| REST API (repositories, issues, pull requests) | API token, or repository, project or workspace access token | endpoints for issues and pull requests are listed; completeness not recorded | default 1,000 requests per hour, scaled 1,000 to 10,000; 100 records per page | support.atlassian.com/bitbucket-cloud/docs/api-request-limits/ |
| Work items | repository users | Jira work items shown in the repository; no Bitbucket export route found | not stated | support.atlassian.com/bitbucket-cloud |
| Wiki | repository users | pages; the edit message becomes a commit message; separate clone not stated | not stated | support.atlassian.com/bitbucket-cloud/docs/create-or-edit-a-wiki-page/ |
| Metadata backup | none | request BCLOUD-19832, status "Gathering Interest" | n/a | jira.atlassian.com/browse/BCLOUD-19832 |
| Data Center | self-hosting administrators | backup, REST API and migration export contents not read | licenses read-only from 2029-03-28 | atlassian.com/blog/announcements/atlassian-ascend |
| Read-only connection | program (Pangea.ai) | owner connects Bitbucket read-only | each program's own process | pangea.ai |
Git Clone and the Repository Size Limits
A repository is cloned over HTTPS or SSH with the Clone button's command, which needs an SSH key or the Git Credential Manager.
Atlassian states that the clone contains the files and metadata Git requires to maintain the changes. Bitbucket Cloud repositories on Standard and Premium plans have a 4.0 GB limit, Atlassian recommends staying under 2.0 GB, and Free plan workspaces have 1 GB in total (read November 17, 2026).
- Over 4 GB, only pushes that undo the latest commits are accepted.
- Atlassian advises Git LFS for large files.
- Git operations are limited to 60,000 requests per hour.
Mercor requires repositories that have "never been public, mirrored or open-sourced" and does not define "mirrored", so the company asks the program whether a Git mirror clone counts before making one.
REST API, Tokens and Rate Limits
Bitbucket Cloud's REST API exposes repositories, issues and pull requests, and its request limits cap how fast a bulk extraction can run.
Atlassian states a default of 1,000 repository data requests per hour, scaled by 10 per hour per paid user above 100 on Standard or Premium workspaces using access tokens, up to 10,000, and says the limits are subject to change.
Pagination returns 10 to 100 records per page, and app passwords are deprecated in favor of API tokens. Which scopes a program receives is the seller's decision, and the comparison of routes sits in buyer connectors vs native exports.
What a Clone Leaves Out: Pull Requests, Work Items, Wiki
A Git clone carries commits, branches and tags, so pull requests, review comments and work items must come from the API or a program, and the wiki is edited through commits.
- Pull requests and issues have API endpoints, and the pages read do not record how complete the returned comments and history are.
- Jira work items appear in a repository's left panel, and the pages read describe no export route for them.
- Atlassian states that a wiki edit comment appears as a commit message and as part of the page history.
- The metadata backup request BCLOUD-19832, created January 24, 2020, shows "Gathering Interest" (re-read November 17, 2026).
Diff limits limit viewing, not export. A buyer that asks for everything in Bitbucket receives what the chosen route holds, so the manifest lists the gaps.
Bitbucket Data Center and the 2029 End of Life
Atlassian announced on September 8, 2025 that Data Center reaches end of life on March 28, 2029, when its licenses expire and become read-only.
New-customer sales end on March 30, 2026, and existing customers' last purchases fall on March 30, 2028 (read November 17, 2026). A self-hosting company exports before a platform move.
What to Remove From Bitbucket Repositories Before Delivery
Bitbucket history carries author names, committed credentials and code the company did not write, so removal starts with the history and ends with the pull request text.
- Credentials and keys committed to files in any branch.
- Author and reviewer names and email addresses in commits and pull requests.
- Pull request and wiki text naming colleagues and customers.
- Third-party and open-source code, checked under the code pages.
Atlassian states that Pipelines variables are set in repository settings, that sensitive values can be marked Secured, and that secrets should not be committed; history keeps what was committed. Names and other personal data follow PII removal and de-identification.
Atlassian's security help page names Snyk as the only security provider available to install with a repository. Secret prevalence and scanner gaps are covered on the GitHub page and under GitLab export and licensing. The outside code question is covered under who owns the code you want to sell.
What Programs Ask For in Bitbucket Repositories
Two programs name Bitbucket, and each publishes different terms, checked on November 9, 2026.
The table lists them alphabetically with the terms that matter for Bitbucket data (as of 2026-11-09; each value is the program's own statement).
| Program | Role | Bitbucket-related records named | Terms that matter for Bitbucket data | Source |
|---|---|---|---|---|
| Mercor, Software Company listing | data buyer | tools named GitHub, GitLab, Bitbucket, Jira, Linear, Confluence | private repositories never public, "mirrored" or open-sourced; 3+ years of commit history from at least 10 engineers; range USD 50,000 to 500,000, more for large codebases or exclusive terms; published 2026-10-17 (checked 2026-11-09) | api.data.mercor.com |
| Pangea.ai code licensing | intermediary (marketplace) | process: "connect GitHub/GitLab/Bitbucket read-only" | non-exclusive; owner keeps ownership; owned private code only; indicative USD 100 per million source-code tokens and USD 40 per million git-history tokens; licensing normally 2 to 16 weeks (checked 2026-11-09) | pangea.ai/code-licensing/companies |
Mercor's USD 50,000 to 500,000 (checked November 9, 2026) is its claim for the whole listing, not a Bitbucket price, and Pangea.ai's rates are indicative claims. No seller-confirmed amount paid for Bitbucket repositories is on record, and the payouts of all programs are compared in AI data partnership payouts and program terms compared.
Licensing Bitbucket Repositories: Benefits, Risks, Downsides and Alternatives
Licensing Bitbucket repositories can earn income from code a company already holds, and it can also expose authors, credentials and third-party code, so not licensing remains a valid result.
Benefits
- Income from existing code, as the programs' own claims.
- Read-only connections keep the working repository in place.
- Pangea.ai states non-exclusive terms and that the owner keeps ownership.
Risks
- Committed secrets and personal data in history.
- Code the company does not own outright, which Mercor and Pangea.ai exclude.
- An exclusive window blocks other buyers; no Bitbucket program publishes one, so the risk depends on the contract.
- Atlassian's processing rights, not read in full.
Downsides
- A clone omits pull requests, work items and the wiki trail.
- API limits slow a large extraction.
- No published price exists for Bitbucket repositories.
Alternatives
- License one repository or an older period only.
- License code together with issue history from Jira, as covered on the Jira page.
- Move to another host first and license after cleanup.
- Do not license at all, as covered in should you sell or license your data.
Bitbucket Among the Code Hosts and the Seller Steps
Bitbucket is one of the code hosts whose terms and export routes decide what a company can license, next to GitHub and GitLab.
The parent page compares export rights and mechanics by source system. The sibling pages are GitHub repositories and their export rights, Jira data export rights for the work items Bitbucket links to, Confluence data export, Linear issue export, open-source license contamination and software companies licensing code.
- Rights audit
- Data inventory
- Customer consent and confidentiality clauses
- Employee data and AI training rules by jurisdiction
- Data manifest
- Security during an AI data sale
- Delivering licensed data
Bitbucket Repositories of a Closed Company
A company that is closing should clone its Bitbucket repositories while the workspace is active, because Atlassian deletes Customer Data after termination in accordance with the Documentation, and the clause states no period.
Atlassian removes a whole repository, not parts, when it removes content. The route is covered in selling data from a defunct company.
The deadlines of other vendors are tracked under exporting company data before shutting down.
Program Pages Named on This Page
The program terms above come from each program's own pages, checked on November 9, 2026.
- Company data partnership programs
- Who buys AI training data
- Mercor data buyer profile
- Pangea.ai code and company data buyer profile
This page is information, not legal advice.
Frequently Asked Questions
How do I export a Bitbucket repository?
A repository is exported with a Git clone over HTTPS or SSH, and pull requests and work items need the REST API. The clone carries commits, branches and tags only.
Does Atlassian use Bitbucket code to train AI?
Atlassian's data contribution page, effective August 17, 2026, names Jira, Confluence and Jira Service Management in its initial scope and does not name Bitbucket.
The page says nothing further about Bitbucket, and the platform settings are covered under opting out of AI training.
Can a company license its Bitbucket repositories to AI companies?
A company can license repositories it owns outright once secrets, personal data and third-party code are dealt with, and two programs name Bitbucket.
Limits are covered under what company data cannot be licensed.
Does a Bitbucket clone include pull requests?
No; a Git clone carries commits, branches and tags, while pull requests are available through the REST API. Their completeness is not recorded in the pages read.