AIPOCH Open-Science v0.28.0: Re-execute and Verify Research Artifacts

AIPOCH Open-Science v0.28.0 adds replayable artifact verification, categorized workspace search, a WSL2 Bash preview, tested Codex upgrades, and stronger privacy defaults.

AIPOCH Open-Science is an open-source, local-first, model-agnostic, self-hosted AI research workbench for reproducible scientific discovery, developed by AIPOCH.

What new capabilities help researchers verify and navigate their work?

The v0.28.0 release lets researchers re-execute a captured artifact version from recorded evidence, compare reproduced files with the original, and retain an exportable verification record. It also adds one categorized search across the workspace, an opt-in WSL2 Bash preview for Windows x64, tested upgrades for the managed Codex runtime, and more explicit controls around remote media and model endpoints.

Re-execute a captured artifact version from a sealed recipe

Provenance is most useful when the recorded process can be tested. AIPOCH Open-Science v0.28.0 can capture version-bound inputs, file generations, an environment lock, lineage, execution evidence, and a replay recipe for a research artifact. A researcher can inspect the evidence selected for a check before the run begins.

When verification starts, Open-Science restores the captured version's sealed recipe in an isolated environment and re-executes it. A check can begin from the captured inputs or from an available intermediate-file checkpoint, and it can run against one artifact version or across multiple versions in the same session.

Execution completion is not treated as proof. The workbench reports comparison outcomes per file, using exact byte checks, bounded comparisons for images and tables, and optional scientific comparison rules. Verification records can be exported, while output retention, size, and deletion controls keep the resulting files manageable.

Python and R dependency analysis covers cross-cell variables and package setup but remains conservative when evidence or an operation is unsupported. The resulting verification record is inspectable evidence for researcher review; it does not establish that a method, interpretation, or scientific conclusion is correct.

Search projects, sessions, messages, files, and literature from one entry point

Research context often becomes difficult to recover when a workspace contains many sessions, uploads, generated outputs, and references. The new global search reaches projects, sessions, message bodies, uploaded files, generated files, and the literature library from one interface.

The All view groups results by category and can load more items within a category. Researchers can filter by scope, date, sender, file format, and literature kind before pagination. Selecting a result opens a contextual detail pane without closing the search, and message results can jump directly to the matching message.

Long messages are fully loaded only when selected. Recent sessions, files, and literature can preview inside the search experience, and file results can link back to their source message. These behaviors make search a navigation layer over the existing research record rather than a separate index with no route back to context.

Keep remote media deferred and require secure remote model endpoints

Remote images, audio, video, and posters embedded in model output can initiate external requests. In v0.28.0, these resources stay deferred until the researcher activates a control that displays the destination hosts. Approval applies to the rendered element and its complete set of URLs, while diagram blocks with embedded remote images are blocked before a request can leave the machine.

Images sent to models are normalized as derivatives with ancillary metadata removed; the original files remain unchanged. Remote model endpoints must use HTTPS, making encrypted transport a required default for non-local connections.

These controls make external data flow more visible, but they do not determine whether a model provider, media host, or connected service is appropriate for sensitive research. Researchers and administrators still need to review destinations, provider policies, credentials, and the content being sent.

Use an opt-in WSL2 Bash preview on Windows x64

Windows researchers who need a Linux-style shell can opt into the WSL2 Bash preview. PowerShell remains the default, and activation requires a readiness-checked WSL2 environment plus version-matched packaged assets.

Commands run through a command-scoped WSL2 sandbox with durable ownership, cancellation, timeouts, restart reconciliation, and tracked cleanup. The feature is a preview for Windows x64, so it should be evaluated in the context of the researcher's local environment and operating requirements rather than treated as the default shell path.

Update the managed Codex runtime through a tested path

The app-managed Codex CLI moves to tested version 0.153.4 with its verified bundled Astra model catalog. Settings now displays the Codex CLI and ACP adapter versions separately and offers an explicit update when an older app-owned pair is detected.

External CLI installations continue to receive manual update guidance. To protect an active run, replacement is refused while an app-launched Codex process is still using the managed runtime target.

What else is new in v0.28.0?

The release also tightens update handling, package access, credential prompts, navigation, and long-session behavior.

Safer update handoff: cancelled downloads do not reappear as ready while a retry is draining, manually downloaded installers are checked for file type, size, and streaming SHA-256 before opening, and a refused install gate leaves a direct retry path.
Operation-scoped package mirrors: configured conda, PyPI, and CRAN mirrors can be authorized for the duration of package management, while mirror URLs with embedded credentials, localhost destinations, or raw IP literals are rejected.
Clearer connector credential decisions: declining a connector credential prompt settles all queued requests for the same credential instead of immediately showing another identical prompt.
Provenance navigation: the Reproducibility entry now appears directly before Review, placing the verification action beside the evidence it checks.
More stable long sessions: queued work and retained history are bounded to keep extended sessions responsive.
Consistent interaction: keyboard and assistive-technology behavior receives a broader consistency pass across the workspace.

Bug fixes that affect daily use

AIPOCH Open-Science v0.28.0 includes targeted fixes across literature, session recovery, agents, compute, provenance, export, and platform dependencies.

Literature: PubMed publication dates and journal abbreviations are preserved during import, and terminal PDF metadata identifiers are checked before use.

Sessions and persistence: optional session fields are compared by value so harmless formatting changes no longer discard edits. Renderer updates, startup recovery, asynchronous cleanup, interrupted data cleanup, and bounded diagnostic logging also receive corrections.

Agents, Skills, and the composer: scoped Codex document loading works again, the Claude Skill loader respects conversation grants, and pasted file references can open before a new session is created.

Notebook and compute: queued compute dispatch continues even when the session catalog fails. Windows R access, kernel startup retry, temporary-directory handling, permission preparation, startup output, and cleanup are repaired.

Verification and provenance: reproducibility recipes remain internally consistent when captured execution evidence is trimmed to its size limit.

Workspace, export, and web: the web interface avoids desktop-only lifecycle APIs, conversation export ignores runtime timestamps during integrity checks, annotation selections trigger during active sessions, and sticky CSV row numbers remain opaque while scrolling.

Platform and dependencies: vulnerable image and tooling dependencies are updated, and source installs can recover stale bundled patches during package installation.

Install or update

AIPOCH Open-Science v0.28.0 supports macOS 12 or later on Apple Silicon or Intel, Linux x64, and Windows 10/11 x64. Download the appropriate DMG, AppImage, Debian package, or Windows installer from the official v0.28.0 release page. Existing installations can update in place where the platform supports it.

Official macOS builds are Developer ID signed and notarized. Windows builds remain unsigned, so SmartScreen may display an “unrecognized app” warning on first launch; confirm that the installer came from the official release page before continuing.

Resources

FAQ

Can AIPOCH Open-Science v0.28.0 re-execute a captured artifact?

It can re-execute a captured artifact version from its sealed recipe and compare the reproduced files with the original. The resulting per-file outcomes are evidence for researcher review, not proof that the research method or conclusion is scientifically valid.

What does artifact verification compare in v0.28.0?

Verification supports exact byte checks, bounded image and table comparisons, and optional scientific comparison rules. A check uses the captured version's recorded inputs, environment lock, lineage, and execution recipe when that evidence is available.

What can the new global search find?

The categorized global search covers projects, sessions, message bodies, uploaded and generated files, and the literature library. Filters, contextual previews, and links back to matching messages help researchers move from a result to its surrounding record.

Is WSL2 Bash now the default shell on Windows?

No. WSL2 Bash is an opt-in preview for Windows x64, and PowerShell remains the default. Activation requires a ready WSL2 environment and version-matched assets.

Does v0.28.0 defer remote media from model output?

No. Remote images, audio, video, and posters stay deferred until the researcher explicitly activates them and can see the destination hosts. Images sent to models are normalized as derivatives, while the original files remain untouched.

Disclaimer

This article is intended for informational purposes only and does not constitute medical advice, clinical guidance, diagnostic recommendations, treatment decisions, or validated scientific conclusions.

AIPOCH Open-Science is a research workflow tool. It does not replace scientific judgment or peer review, and researchers remain fully responsible for evaluating the accuracy, completeness, and appropriateness of generated outputs. All outputs require independent verification and expert interpretation before use in any research context.

References and external links in this article are provided for informational purposes. AIPOCH does not endorse and is not responsible for the content of third-party sources.