Release notes · v0.37.0
Released on 2026-10-11 UTC. Official release.
Run without a desktop window
The desktop, browser, CLI and Notebook now share an ordinary Node backend. open-science start no longer launches a hidden Electron host. Official releases include standalone CLI archives with Node and the backend bundled; no separate Node installation is needed for these archives. Desktop backends choose a free port when none is specified.
Use terminal setup for installation and Codex login. Read service ownership and shutdown before closing a desktop client or stopping a shared backend.
More database evidence
The registry now contains 36 Connectors and 361 operations: 35 data-source families with 359 operations, plus two offline Molecule operations.
| Entry | What changes | Guide |
|---|---|---|
| ENCORI | Ten operations for miRNA targets, RNA–RNA and RBP interactions, ceRNA networks, motif scans, binding sites, reference tables and bulk files | Setup and result boundaries |
| ChEMBL | Assay details and bioactivity pagination beyond the first 1,000 records | Assay context and pagination |
| ClinVar | Individual SCV submission evidence for a VCV record | Compare submissions |
Keep organism, assembly, filters and evidence types with the results. Assay confidence is about target assignment; a ClinVar submission is not an independent clinical diagnosis. The operation reference provides exact fields and a downloadable registry.
PDF reading and Notebook execution
PDF translation improves native reading order and nested text labels. Structure extraction keeps figures and table records with their source elements. Continue checking figures and tables and translated pages against the original.
Ready default Notebook runtimes avoid redundant setup approvals. Mixed Conda/pip environments and micromamba build metadata are captured more accurately, alongside input and callback dependencies. These records support rerun verification; they do not themselves establish reproducibility.
Recovery and platform fixes
- Import into a new project fixes a
.sciencecatalog-adoption failure. Imported sessions remain read-only; use Fork to continue analysis. - Smart collections retry the failed batch while retaining completed work. See screening a collection.
- Plan approval/dismissal and Codex streamed tool arguments recover more reliably. Auto mode avoids redundant approval for supported inspection operations.
- Windows fixes cover Python paths containing spaces or parentheses, selected backend profiles and Specialist handoffs. Linux credentials avoid repeated Secret Service probes after the first secret operation.
- PDB rejects malformed search responses and preserves incomplete-pagination status; an error must not be read as zero hits.
- Home adds first-project guidance. Headless and SDK execution can resolve bundled Notebook resources again.