AIPOCH Open-Science v1.0.3

A connector-and-literature expansion: seven new Scientific Connector tools spanning imaging, pathways, lipids, structures, and cancer models; a Literature library upgrade with Semantic Scholar citation discovery, supplementary attachments, drag-and-drop import, and smart-screening CSV export; session-package export and import from headless CLI and local Web sessions; an opt-in content-first layout for translated PDFs; notebook folder access restrictions; and a broad round of Windows, sandbox, session-recovery, and PDF fixes.

AIPOCH Open-Science is an open-source, local-first, model-agnostic AI research workbench for scientists and researchers. It enables reproducible, inspectable research across models with scientific AI agents, Python and R execution, scientific data connectors, and cross-platform support for macOS, Windows, and Linux.

v1.0.3 follows v1.0.2 with a wide scientific-coverage expansion and a deep reliability pass. The Scientific Connector catalog gains seven new tools — The Cancer Imaging Archive, KEGG, OmniPath, LIPID MAPS, PDBe validation with UniChem mapping, DepMap and CCLE cancer models, and PDB structure similarity search — all behind default-off discovery with persisted per-source opt-in. The Literature library adds Semantic Scholar citation discovery, multi-source supplementary attachments with download progress, drag-and-drop import, and spreadsheet-ready Smart Screening export. Session packages can now be exported and imported from headless CLI and local Web sessions, translated PDFs can reflow into an opt-in content-first layout, and notebook folder access restrictions keep execution inside folders you authorize. Kimi K3.1 Pro models join the provider catalog, and a broad set of fixes hardens Windows, the notebook network sandbox, session recovery, and PDF handling.

✨ Highlights

• Seven new Scientific Connector tools span imaging, pathways, lipids, structures, and cancer models. New tools reach The Cancer Imaging Archive (TCIA) imaging collections and KEGG pathway queries (both opt-in), OmniPath molecular regulation, exact LIPID MAPS compound lookup, PDBe structure validation with UniChem cross-mapping, DepMap and CCLE cancer cell-line queries, and PDB structure similarity search — all with default-off discovery and persisted per-source opt-in. (#3760, [#3799], [#3805], [#3808], [#3815], [#3818], [#3828])
• The Literature library becomes a full research inbox. Semantic Scholar citation discovery surfaces a paper's references and citing papers, multi-source supplementary attachments download with progress and preview (including Word documents), PDFs and bibliography files import by drag-and-drop, and Smart Screening results export as a spreadsheet-ready CSV. (#3791, [#3787], [#3776], [#3822])
• Session packages work from anywhere. .science export and import now run from headless CLI and local Web sessions, not only the desktop app. (#3801)
• Translated PDFs gain an opt-in content-first layout, and notebooks get folder access restrictions. Translated editions can reflow body text and captions into single-column pages while keeping images, tables, formulas, and bibliography original, with source-to-output page mappings for comparison. Notebook folder access restrictions, on by default, keep notebook and shell execution inside folders you authorize. (#3637, [#3800])
• A broad Windows, sandbox, session-recovery, and PDF reliability round. The open-science protocol registers correctly on Windows again, sandbox network approvals persist across restarts, session recovery survives model failures and can delete damaged conversation files, and PDF structure analysis excludes clipped text and wrapped figure references. (#3777, [#3830], [#3740], [#3804], [#3795])

🚀 New Features

• Opt-in TCIA (cancer imaging collections) and KEGG (pathways and drugs) connector tools. (#3799)
• OmniPath queries across molecular regulation and signaling interactions. (#3805)
• Exact LIPID MAPS compound lookup by identifier or name. (#3808)
• PDBe structure validation with UniChem mapping to connected chemical databases. (#3815)
• DepMap and CCLE cancer model queries for cell-line dependency data. (#3818)
• PDB structure similarity search starting from a structure identifier. (#3828)
• Semantic Scholar citation discovery in Literature, with an optional API key for higher rate limits. (#3791)
• Multi-source supplementary attachments with per-file download progress, selected-file archive extraction, and previews by file type including Word documents. (#3787)
• Drag-and-drop import of PDFs and bibliography files into the Literature library, plus attachment downloads from the reference detail view. (#3776)
• Smart Screening results export as CSV, covering all candidates with final verdicts, manual decisions, reasons, and model probabilities. (#3822)
• .science session package export and import from headless CLI and local Web sessions. (#3801)
• Content-first layout option for translated PDFs, off by default and remembered per translation edition. (#3637)
• Notebook folder access restrictions, enabled by default for new setups, confining notebook and shell execution to authorized folders. (#3800)
• Local-files tooling can list authorized directories on demand, so agents reuse existing folder grants instead of asking again. (#3767)
• Kimi K3.1 Pro provider models are available for selection. (#3835)

🔧 Improvements

• Session package transfers move more data by default, and import skips repeated verification passes, so large archives finish sooner. (#3782, [#3780])
• PDF annotation and structure analysis are faster on large documents. (#3825, [#3831])
• Specialist switches show progress while the new specialist loads. (#3778)
• Settings editors lock their inputs while saving and protect unsaved drafts across Memory, Tags, and credentials screens. (#3781, [#3849])

🐛 Bug Fixes

• Windows: the open-science protocol registration is restored, the notebook host rebuilds before startup in development setups, and session packages rehash content correctly on Windows. (#3777, [#3843], [#3829])
• Session recovery now survives model failures, and damaged conversation files can be deleted instead of blocking recovery. (#3740, [#3804])
• Specialist switching recovers from certain provider rejections without losing the conversation, and capability failures are distinguished from unavailable bindings. (#3772, [#3796])
• The composer switches correctly between stop and queued send. (#3768)
• Session packages: research code no longer trips credential false positives, and export resumes properly after a sensitive-content acknowledgement. (#3817, [#3846])
• Notebook network sandbox: approvals persist across restarts, wildcard rules and URL input are accepted, native proxy routing is enabled, and filesystem and proxy boundaries stay enforced. (#3830, [#3826], [#3827], [#3841])
• Notebook analysis preserves lineage across chained scientific effects, signed shell exit codes survive, protection probes no longer slow session startup, and shell searches stay inside the filesystem sandbox. (#3812, [#3851], [#3839], [#3788])
• PDF handling excludes clipped text and wrapped figure references from structure analysis, hardens figure, table, and reader ownership, and simplifies single-page previews. (#3795, [#3814], [#3816])
• Desktop recovery resolves reproduced title, retry, and protection defects. (#3823)
• Claude subscription probe arguments are preserved in settings. (#3837)
• Unrelated services are excluded from runtime process cleanup. (#3797)
• Browser imports preserve retries across suspension. (#3770)

📦 Install

Desktop client. Requirements: macOS 12+ (Apple Silicon or Intel), Linux x64 or ARM64, or Windows 10/11 x64. On first run, the onboarding wizard checks the environment and can install and configure an app-managed agent runtime. Once installed, the app updates itself in place.

Download the appropriate package from the Assets section below, or from the official download page. On macOS, the desktop app is also installable via Homebrew: brew install --cask open-science.

Platform	Package
macOS (Apple Silicon)	DMG for ARM64
macOS (Intel)	DMG for x64
Linux	AppImage or Debian package for x64 / ARM64
Windows	Code-signed installer for x64

Command line via Homebrew (macOS and Linux). The standalone CLI and local Web UI live in the official tap — bundled Node runtime, no separate Node.js needed, and upgrades through the normal brew flow:

brew install aipoch/tap/open-science
open-science start --no-open
open-science url

Command line via npm. With Node.js 22.13+, install the CLI (and Task SDK) globally; updates then follow your normal npm upgrade flow:

npm install --global @aipoch/open-science
open-science --help

Command line standalone (all platforms, no toolchain). The Assets below also include standalone CLI archives — macOS and Linux arm64/x64 tarballs and a Windows x64 zip — with a bundled Node runtime, so they install and run without Homebrew, npm, or a local Node.js. Linux packages additionally arrive as standalone .deb and .rpm installers.

macOS — first launch. Official release builds are Developer ID signed and notarized by Apple, so they open like other trusted applications. A locally built copy is notarized differently and may require approval through macOS Privacy & Security.

Windows — first launch. Official release builds are code-signed; Windows recognizes the publisher and no SmartScreen warning appears. Verify that the package came from the official release page before continuing.

Build from source instead:

npm install
npm run build:mac   # or: build:linux / build:win

🧭 What's in this release (maturity)

• ✅ Implemented: everything shipping in v1.0.2, plus the seven new Scientific Connector tools, the Literature library upgrade (citation discovery, supplementary attachments, drag-and-drop import, Smart Screening CSV export), session packages from headless CLI and local Web sessions, the content-first translated-PDF layout, notebook folder access restrictions, Kimi K3.1 Pro models, and the fixes described above.
• 🚧 Partial: provider choice remains constrained by the active framework's endpoint compatibility; remote compute covers direct SSH and Slurm (cloud-GPU submission is not built yet); WSL2 Bash remains an opt-in Windows x64 preview; the skills commons ships marketplace discovery but not cross-machine forking or user-facing version pinning; reproducibility ships replayable per-version verification, session replay, and RO-Crate packaging, while solver-exact equivalence remains open; review is opt-in and record-scoped.
• 🗺️ Roadmap: a unified model gateway, cross-machine skill forking and version pinning, cloud-GPU execution, stronger sandboxing and credential isolation, and collaborative research workflows.

🐢 Known Limitations

• Replay reproduces recorded steps, it does not re-execute them. Solver-exact equivalence guarantees remain open; replayable verification remains the deterministic check, and starting a check from the web client currently requires the desktop app.
• Imported .science sessions are read-only. They stay inspectable and referenceable, and forking them into a writable continuation remains available; execution and continuation remain disabled on the import itself.
• PDF annotations are bound to the file version they annotate. Notes follow the file across projects and sessions where that version is shared, but they do not sync across machines.
• Bookmarks belong to a single session. They do not follow branch switches, join .science package exports, or sync across machines — forking copies them into the new session with fresh identities.
• WSL2 Bash is a preview. It is opt-in on Windows x64 only; PowerShell remains the default shell runtime.
• Remote compute covers direct SSH and Slurm. Cloud-GPU submission is not built yet.
• Editable artifacts cover text formats. Binary formats stay read-only, and editing always publishes new versions.
• Network sandboxing covers the app's notebook and compute runtimes, not the whole system. On Windows, the boundary applies only after the sandbox's one-time administrator setup.
• Provider choice is per framework, not one unified gateway. The available protocol depends on the selected agent backend, and remote endpoints must use HTTPS.
• Code reconstruction is LLM-generated. It does not replace deterministic reproduction; replayable verification is the deterministic check.
• The reviewer is opt-in and record-scoped. It does not replace domain-specific validation of citations, units, statistics, or methods.
• No local GPU compute backend.
• No multi-user real-time collaboration.

🙏 Acknowledgements

Thanks to @ewen-poch, @nasus2002, @roxi3906, @thxuan, @justemu and everyone in Discord, X, and Discussions.

---

Full Changelog: https://github.com/aipoch/open-science/commits/v1.0.3