Skip to main content

Changelog · v0.32.0

Released September 21, 2026. This release adds persistent PDF reading notes, sequence similarity search and more ways to discover public sequencing and proteomics data.

Official release · Changes since v0.31.1

PDF annotations and document notes

Highlight a passage, mark an area, or add a page or document note. Notes belong to the managed file version and can carry comments and tags. Find them again in the notes view or global search, and export Markdown, CSV or a separate annotated PDF. The original PDF bytes stay unchanged.

Library attachments share their notebook wherever the same managed version is used; project uploads and artifacts share it across sessions in the owning project. See the illustrated PDF notes guide for marking, searching and exporting. These notes are distinct from private session bookmarks.

Sequence and omics discovery

CapabilityWhere to start
UniProt discoveryFind entries by gene, protein phrase and organism, then retrieve the selected sequence. Protein search
NCBI BLASTSubmit a public nucleotide or protein sequence, retain its RID, follow the polling interval and retrieve the completed report. Sequence search
ENA discoveryCombine organism, library strategy and keyword filters. List archive FASTQ and original submitted files separately. Discovery and file inventory
PRIDE file listingsPage through project file metadata and public locations without downloading the data. File inventory

The operation reference now covers 251 data-source operations, plus two Molecule operations, across 24 Connectors. Mitochondrial gnomAD queries accept gnomad_r3 and gnomad_r4 datasets.

Models and automation

  • Custom HTTP service connects capability selection to a TypeSafe-compatible classifier. It has its own endpoint, model and key; it does not change Main. See classification models.
  • StepFun Step-5 Preview adds China and Global regions. Select the endpoint region in provider settings.
  • Local PDF parsing models can use approved, integrity-checked download mirrors when the primary source is unreachable. See PDF extraction.
  • CLI and SDK tasks can use permission-prompts none to decline requests that would wait for a person. This preserves existing approval rules; it does not grant permission automatically.

The release also adds the complete RO-Crate 1.1 packaging profile in the export backend: it includes a verified artifact version and exact input bytes, checks declared sizes and checksums, deduplicates identical content and rejects conflicts. Existing .science handover and per-version reproducibility remain the desktop workflows described in this Wiki.

Fixes and interaction changes

Side Chat keeps conversations and queued advice for the current app run. Switching views or reloading the interface can reconnect to them. Confirming Close Side chat removes that discussion, and fully restarting the app clears remaining Side Chats. Messages already delivered to Main remain in Main.

Other changes improve startup and long conversations, linked-PDF reading, background compute delivery and cancellation, per-session OpenCode tool isolation, PDF text selection, settings search, file focus and recovery alerts. See the official release for the complete list. The preceding release is v0.31.1.