Releases: synthetic-sciences/openscience
Release list
v2.0.151
More reliable research sessions
- Keep GPT-6 Sol and GPT-6.1 Sol available through connected ChatGPT/Codex subscriptions, with supported reasoning controls and Fast mode for 6.1. GPT-6 Sol is removed from the Ace roster.
- Stop unattended runs from inventing work after a greeting.
- Preserve tool availability across retries, restore sessions containing malformed todo history, and keep pagination complete when sessions share a timestamp.
- Prevent stale Codex sign-in timers from clearing newer attempts and keep provider web-search events consistent.
Workspace and desktop
- Recover the newest verified pending desktop update.
- Preserve unsaved-file warnings when new edits arrive while a confirmation dialog is open.
- Restore the live Modal job panel and retain balance-recovery guidance across gateway response formatting changes.
- Preserve skill acronym casing and display escaped Markdown safely when sanitization is unavailable.
Datasets and scientific helpers
- Create Hugging Face datasets privately by default; use
--publicto publish a new dataset publicly. Upload failures now return a failing exit status. - Validate question-answering, completion, and union fields. SQL table substitutions preserve quoted text, comments, and unrelated identifiers.
- Restore BRENDA fallback enzyme queries, correct filtered market-visual dry runs, and report accurate zero-TMB results when every variant is filtered out.
Runtime fixes
- Generate usable patch previews with correct line counts, blank lines, and end-of-file handling.
- Preserve fetched page text around skipped and embedded HTML elements.
- Diagnose corrupt configuration and process ledgers, recover kernels around malformed saved records, and reject non-exact locked Python requirements.
- Return consistent MCP status maps, honor zero session limits, handle HTTP range units correctly, and tolerate diagnostic lines around SSH control responses.
Thanks to ANIRUDDHA ADAK (@aniruddhaadak80) for the community fixes incorporated into this release.
Full changelog: v2.0.150...v2.0.151
v2.0.150
Models and provider setup
- Add OpenAI GPT-6.1 Sol with supported reasoning controls and Fast mode.
- Configure custom model Base URLs and OpenAI protocol selection in Settings.
- Removing a provider key disconnects its saved connection while retaining endpoint and protocol preferences.
- Improve research-search setup guidance and Chinese settings and connection messages.
Terminals and Claude Code
- Open new terminals in the selected working folder and load the user’s login shell and installed commands.
- Give user-operated terminals normal local access while preserving the research agent’s sandbox.
- Launch the installed Claude Code CLI from the Terminal panel with its existing login and session history. This is an interactive terminal session; the research-chat model remains independently configured.
Remote workspaces and scientific connectors
- Open a trusted remote runtime through an existing SSH tunnel in an isolated desktop window. Files, events, and terminals stay on the selected server. The tunnel must already be running; authenticated-runtime sign-in and automatic SSH management remain follow-up work.
- Add keyless BioSimulators registry search and version-specific simulator records.
- Search every BioSimulators model format by its name, including Kappa and Virtual Cell Markup Language. Thanks to Richard Sundby Anderssen (@randsu) for #911.
Windows desktop builds continue to use verified, timestamped signatures from Inkvell Inc. macOS desktop builds are signed and notarized.
Full changelog: v2.0.149...v2.0.150
v2.0.149
Windows signing
- Sign the Windows desktop installer, application, bundled CLI, and native libraries with Microsoft Artifact Signing.
- Verify trusted, timestamped signatures from Inkvell Inc. before publication.
This patch includes the fixes and improvements from v2.0.148.
Full changelog: v2.0.148...v2.0.149
v2.0.148
Fixes and improvements
- Correct provider pricing at tier boundaries, including Fast pricing, and recognize local inference endpoints without a URL scheme.
- Improve command approval patterns, reject invalid shell timeouts and empty fuzzy edits, and make capped search results deterministic.
- Preserve complete output preview limits and report truncated text accurately.
- Fix file previews, nested-bracket image descriptions, quoted empty CSV cells, and large mass-spectrometry scans.
- Respect configuration precedence and push the branch selected in the workspace to its own tracked remote.
- Improve scientific helpers for local codon GC targets, CFSE input validation, enzyme transformation matching, and unnamed GPU devices.
- Fix queue and timer cleanup, duration labels, file-path helpers, and attachment-size display.
- Update editor, molecular viewer, schema-conversion, and development dependencies; keep CodeMirror extensions on a single state version.
- Update the website's benchmark note.
Community contributions
Thanks to ANIRUDDHA ADAK (@aniruddhaadak80) and yi111 (@Yi-111-a) for the reports and fixes consolidated in this release, and to Dependabot (@dependabot) for dependency updates. Contributor credit is retained in the merged commits.
Full changelog: v2.0.147...v2.0.148
v2.0.147
Core
- Send OAuth refresh tokens and client secrets only to the issuing authorization server
The Windows desktop installer is unsigned while Microsoft Artifact Signing setup is incomplete.
v2.0.146
Core
- Break glob mtime ties by path (#809) (ANIRUDDHA ADAK)
- Read the installed-skill entry list on any platform (#803) (ANIRUDDHA ADAK)
- Treat --port=N as an explicitly passed port (#795) (ANIRUDDHA ADAK)
- Refuse an empty question list (#793) (ANIRUDDHA ADAK)
- Report a separator-only fetch selection as not found (#791) (ANIRUDDHA ADAK)
- Do not count a trailing newline as a truncated line (#789) (ANIRUDDHA ADAK)
- Rebase onto current main (#782) (ANIRUDDHA ADAK)
- Read the code search reply in the format the request asked for (#785) (ANIRUDDHA ADAK)
- Tell a missing import file from an unreadable one (#815) (ANIRUDDHA ADAK)
- Skip comments and blank lines in pinned requirements (#813) (ANIRUDDHA ADAK)
- Say how many skill matches a query left out (#811) (ANIRUDDHA ADAK)
- Never let a synthetic run take a journal ordinal (#784) (ANIRUDDHA ADAK)
- Let openscience.json win over openscience.jsonc globally (#821) (ANIRUDDHA ADAK)
- Require a command on a local MCP server entry (#819) (ANIRUDDHA ADAK)
- Read a file:// folder link on Windows (#817) (ANIRUDDHA ADAK)
- A malformed patch is rejected instead of reported as applied (#826) (ANIRUDDHA ADAK)
- Count reasoning tokens once in the session spend (#787) (ANIRUDDHA ADAK)
- Classify an abbreviated long option like the flag it reaches (#781) (ANIRUDDHA ADAK)
- Keep the indentation a fuzzy match spans (#780) (ANIRUDDHA ADAK)
App
- Finish guarding browser storage reads (#827) (ANIRUDDHA ADAK)
- Measure a match across the original line, not the folded one (#805) (ANIRUDDHA ADAK)
- Key the transcript snapshot guard by the stream directory (#825) (ANIRUDDHA ADAK)
SDK
- Percent-encode non-ASCII directory in the v1 client header (#829) (Mikhail R. Nemerov)
Thank you to 2 community contributors:
- ANIRUDDHA ADAK:
- Break glob mtime ties by path (#809)
- Read the installed-skill entry list on any platform (#803)
- Treat --port=N as an explicitly passed port (#795)
- Refuse an empty question list (#793)
- Report a separator-only fetch selection as not found (#791)
- Do not count a trailing newline as a truncated line (#789)
- Rebase onto current main (#782)
- Read the code search reply in the format the request asked for (#785)
- Finish guarding browser storage reads (#827)
- Tell a missing import file from an unreadable one (#815)
- Skip comments and blank lines in pinned requirements (#813)
- Say how many skill matches a query left out (#811)
- Never let a synthetic run take a journal ordinal (#784)
- Let openscience.json win over openscience.jsonc globally (#821)
- Require a command on a local MCP server entry (#819)
- Read a file:// folder link on Windows (#817)
- A malformed patch is rejected instead of reported as applied (#826)
- Count reasoning tokens once in the session spend (#787)
- Measure a match across the original line, not the folded one (#805)
- Classify an abbreviated long option like the flag it reaches (#781)
- Keep the indentation a fuzzy match spans (#780)
- Key the transcript snapshot guard by the stream directory (#825)
- Mikhail R. Nemerov:
- Percent-encode non-ASCII directory in the v1 client header (#829)
The Windows desktop installer is unsigned while Microsoft Artifact Signing setup is incomplete.
v2.0.145
No notable changes
The Windows desktop installer is unsigned while Microsoft Artifact Signing setup is incomplete.
v2.0.144
Core
- Allow non-secret query parameters in remote MCP URLs; say when a local MCP server was refused
- Move codesearch to Exa web search and surface server errors
- Add nmr-compound-inference skill (#169) (Yaobin Wu (Robin))
- Prefer layers for single-cell count preservation (#116) (Cmf)
- Do not read a failed enumeration as an empty project store (#783) (ANIRUDDHA ADAK)
- Fix(patch): insert an added-only hunk where its header says (#779) (ANIRUDDHA ADAK)
- Let the nearest .openscience directory win (#799) (ANIRUDDHA ADAK)
- Send the UCSC start coordinate 0-relative (#797) (ANIRUDDHA ADAK)
- Require kernel 3.10, not 5.1, for the bundled runtime (#778) (ANIRUDDHA ADAK)
App
- Translate the Models and Connectors settings (#774)
- Read the localStorage global defensively (#771) (ANIRUDDHA ADAK)
- Read a Windows worktree folder name (#807) (ANIRUDDHA ADAK)
Thank you to 3 community contributors:
- Yaobin Wu (Robin):
- Add nmr-compound-inference skill (#169)
- Cmf:
- Prefer layers for single-cell count preservation (#116)
- ANIRUDDHA ADAK:
- Read the localStorage global defensively (#771)
- Do not read a failed enumeration as an empty project store (#783)
- Fix(patch): insert an added-only hunk where its header says (#779)
- Read a Windows worktree folder name (#807)
- Let the nearest .openscience directory win (#799)
- Send the UCSC start coordinate 0-relative (#797)
- Require kernel 3.10, not 5.1, for the bundled runtime (#778)
The Windows desktop installer is unsigned while Microsoft Artifact Signing setup is incomplete.
v2.0.143
Core
- Launch local study runs from a study folder in Project files
The Windows desktop installer is unsigned while Microsoft Artifact Signing setup is incomplete.
v2.0.142
Core
- Let the agent read a PDF attached to a chat message
The Windows desktop installer is unsigned while Microsoft Artifact Signing setup is incomplete.