Skip to content

Babel Desktop 0.1.3

Choose a tag to compare

@github-actions github-actions released this 14 Oct 21:34
· 385 commits to main since this release
abc05bb

Babel Desktop 0.1.3 is the Windows x64 desktop app for the Babel coding-agent harness. It bundles the CLI, prompt resources, production dependencies and Node runtime. Provider setup, enabled/default model selection, project files, approvals and persisted history use the real CLI-backed workspace.

Download Babel-Desktop-Setup-0.1.3-win-x64.exe from Assets below. The ZIP is a portable alternative; the CLI tarball is included for reproducibility.

  1. Run Setup.exe and open Babel Desktop.
  2. Start Docker Desktop with Linux containers. In Babel Setup, prepare the recommended image or validate an existing local toolchain image.
  3. Open a project. In Providers and models, select a provider, add its API key, enable models and choose a default.
  4. Send a task in Chat. Commands and file changes follow Babel's approval and execution rules.

The installer and portable bundle are unsigned. Verify the installer against its matching .exe.sha256 file. Setup SHA256:

2cc94aee338edfa34059ad02aa508cd92062051741bff2eb8e7cec42b7aeffe1

Existing profiles are preserved. API keys are saved in a private profile .env file (plaintext), rather than an OS keychain. Plan and Deep currently require OpenRouter; other supported providers offer Chat. Custom project tools may require a custom sandbox image. Automatic Desktop updates are unavailable.

Windows setup guide · Source commit · Desktop workflow fixes · Sandbox readiness fixes

Qualification: post-merge CI passed, including installer lifecycle; release build binds the annotated tag and source. Downloaded assets passed independent checksums, all 10,234 ZIP manifest files and build/installer provenance checks. The actual downloaded Setup.exe passed isolated install, registration/shortcut, 12 native UI checks, CLI launch and uninstall. The release payload also passed provider key entry, enabled/default model persistence, blank-key reuse and restart checks without inference. These checks do not establish general live-model coding reliability.