VOID Mind v0.2.0-alpha.1

VOID Mind v0.2.0-alpha.1

An early standalone intelligence workspace. Manual validation follow-ups remain open and are listed below; publication does not mean those checks passed.

VOID Mind becomes a standalone intelligence workspace with Home, conversations, persistent Workspaces and explicitly selected Knowledge.

Included

• Continue with ChatGPT through the official direct connection, without requiring a separate Codex installation. An OpenAI API key is an alternative connection; catalogs and provider preferences remain separate.
• Streaming conversations, Stop with preserved partial content, continuation and restart persistence. New conversations enter history only after the first message. Titles remain deterministic and can be renamed.
• Workspaces with descriptions, instructions and model preferences. Conversation overrides take precedence over workspace and global preferences.
• Global model indicators and reasoning controls follow the configured provider/model, including after catalogs refresh, instead of borrowing another provider's cached selection.
• Explicit Markdown/text imports into app-managed storage, source previews and visible context selection. Removing imported content preserves the original file. This is manual context, not semantic retrieval or automatic memory.
• Urbanist/Barlow typography, shared Base UI controls, VOID focus/scroll treatments, integrated window controls and branded installer assets. Disabled send buttons retain the normal arrow cursor; the browser authorization callback includes the VOID favicon.
• Optional compatibility with existing Codex-bound conversations, including bounded executable discovery, cooperative process shutdown and a restricted neutral workspace.

Availability and validation limits

ChatGPT availability and usage depend on the connected OpenAI account. API-key requests are billed separately by the provider. No extra AI request is made to name a conversation.

Real ChatGPT login, catalog, streaming, Stop, continuation, explicit workspace/source context and restart recovery have been exercised. Real API-key QA remains in issue #2. Remaining manual validation is tracked in issue #3 for the final Windows installer/UI, issue #4 for real credential renewal/remote revocation, and issue #5 for legacy Codex lifecycle/authentication and Linux installation. Those checks are not claimed as passed.

Real llama.cpp/GGUF inference validation was deferred by the maintainer to issue #1; the provider remains secondary and its automated regression is retained. The Python laboratory remains in standby. Qwen is not integrated, and no training or new model benchmark is included.

The tested Windows Codex runtime/profile denied a synthetic read outside its neutral workspace while allowing the positive control. This is not a claim of exhaustive filesystem/network enforcement on every platform. Direct conversations expose no tools. Linux package acceptance currently uses extracted Debian/AppImage applications rather than an OS package-manager installation or FUSE mounting.

See the completion audit, validation receipts and connection details for the exact evidence and remaining gates.

Automated validation

The application source at aaa6d8d passed Windows/Linux validation: 67 JavaScript tests in 20 files, 38 Windows and 37 Linux native tests, TypeScript, lint, frontend/Tauri builds and packaging. Extracted Debian passed eleven native acceptance checks; extracted AppImage passed twelve, including reduced motion. Six Python laboratory tests and Ruff also passed locally. Subsequent release-preparation edits affect documentation only.

Release run 40901467460 also passed validation and packaging on both platforms at tag commit 1fefd1a, then published three packages and three checksum files. Public asset digests match the downloaded workflow files. Final installed/manual checks remain in the issues above.