RL ATLASA startup winding down may still own useful code and a record of how its team worked. Repositories, tickets, reviews, documentation, and internal workflows can help AI teams study or test agents on realistic tasks. The opportunity is time sensitive: access to hosted tools and the people who understand the archive can disappear during shutdown.
Preserve the material first. Then decide whether a buyer exists, what can legally be transferred, and which transaction structure fits the company.
Save the history while systems are available
Export connected records before subscriptions end. A final repository ZIP may omit the issues, pull request discussions, and project decisions that make the engineering history intelligible.
Create a controlled archive of the relevant systems:
| System | Records to preserve |
|---|---|
| Code hosting | Repositories, branches, commits, pull requests, reviews, and issues |
| Project tools | Tickets, statuses, discussions, and links to code |
| Documentation | Design decisions, runbooks, and incident reports |
| Support and operations | Workflows and outcomes the company owns and may use |
| Internal databases | Schemas and records that may help explain the product or processes |
Record what each export contains, its date range, format, and original system. Keep the relationships between records where possible. Put access controls around the raw archive. Preservation is not permission to disclose the material; rights and privacy review come next.
Decide which assets belong in the package
Code with working tests may support coding tasks. Tickets linked to verified fixes can show how requests became changes. Operational records may show a complete workflow and its result. Buyers can assess these as separate assets or as one connected package.
Exclude material the company cannot transfer or license. That may include customer data, employee personal information, credentials, third-party confidential documents, and code with unresolved ownership. Review contributor and contractor IP agreements, customer contracts, open-source dependencies, security interests, and any required company approvals. The company should have a clear chain of authority when it signs a transaction. Seek qualified legal review for the actual sale or license documents.
Prepare a small, sanitized example of the retained material. Do not send a full workspace archive merely to test buyer interest.
Sale and license solve different problems
An asset sale can provide a fixed payment and transfer ownership of the specified assets. It may be practical when the company will dissolve and nobody will remain to manage licensing. A license can preserve ownership and permit more than one transaction, but someone must hold the rights, administer agreements, and handle ongoing obligations after shutdown.
Compare offers using the net proceeds, time to payment, scope of transferred rights, responsibility for preparing the archive, and obligations that survive closing. A large nominal price can be less useful if the transaction cannot complete before dissolution or requires work the team cannot perform. Public deal figures describe particular assets, not a standard price for a failed startup.
Our source-code guide covers what makes a repository useful to a buyer. The AI data licensing guide explains the usage terms that matter if the company or a successor will retain ownership.
Run a focused process
Assign one person to own the archive and one to track rights and approvals. Export the systems, make an asset inventory, mark exclusions, and prepare a representative sample. Approach buyers or marketplaces with a precise description and a realistic closing timeline. DataVendor is one marketplace that lists codebases and datasets, but a listing alone is not a sale.
Do not erase the original records until retention duties and transaction needs are understood. Do not assume that every part of the archive has equal value. A smaller package with clear ownership, linked context, and a usable delivery format is easier for a buyer to assess than an unexplained dump of everything the company ever stored.