Your workbench, ready to discover.

DeloresAI accepts free applications with meaningful AI, machine learning, scientific, or optimization functionality. Your GitHub repository remains the source of truth for source code and release assets.

From source to marketplace

1. Choose a public or private GitHub repository. Public projects must declare their license. Private projects must authorize source inspection and free marketplace distribution.
2. Build a useful workbench that stays free to acquire and use. Disclose external API, data, compute, and model costs.
3. Produce native installers on the appropriate operating system. Keep lockfiles, source commit, notices, and software bill of materials with your release.
4. Upload installers to a versioned GitHub Release. Source archives alone are not installers.
5. Create and verify your DeloresAI account. Open the publisher workspace and connect your GitHub identity.
6. Authorize the central DeloresAI GitHub App for specific repositories. Repository access and your GitHub identity are checked separately.
7. Select the authorized repository and release. Map every installer to its operating system, processor, format, and minimum requirements.
8. Generate your manifest, review its privacy and external-cost disclosures, and submit.
9. Read the automated decision and findings for each platform. Correct a finding in a new release or retry when unavailable verification infrastructure becomes ready.
10. Future release events become candidates automatically. Each target advances only after the required checks pass.

Build with your preferred framework

Your app does not have to use Tauri. Electron, native Rust/C++, and frozen Python can produce package families described by the manifest. Actual admission depends on the verification and installation adapters available for that target.

Build Windows packages on Windows, macOS packages on macOS, and Linux packages on the supported Linux distribution. Use deterministic dependencies, exact version tags, final-byte checksums after signing, and short artifact retention.

A failed target doesn’t erase your release

For example, a Windows update can pass while Linux remains on its previous eligible version. Findings say exactly which target failed or remained incomplete. A prior approved, unchanged release stays available when a routine rescan is late or unavailable. Its original check dates remain visible, and downloads still need current authorization.

What publishers never need to provide

You do not need to paste a personal access token, create your own DeloresAI integration keys, or share cloud-provider credentials. Workbench users’ API keys belong in the workbench, not in the marketplace.