Skip to content

v0.6.0

Choose a tag to compare

@github-actions github-actions released this 22 Jul 15:58
· 3467 commits to main since this release
9eba472

The remote-compute and SDK release: connect a remote SSH host for notebook execution, get desktop task notifications, drive the workbench programmatically from a zero-dependency Node.js task SDK, and export any session as a .ipynb — while the model layer gains Volcengine Ark and per-model reasoning-effort control.

Open Science is an open-source, model-agnostic AI workbench for scientific discovery — a self-hosted desktop app pairing a plan-and-execute agent with a persistent, sandboxed compute kernel and durable project/session storage.

This is a minor release: no breaking changes. It advances two open Roadmap phases — Phase 4 (remote compute + async notifications) and Phase 1 (CLI/SDK surface) — alongside the standard provider and reliability work.

✨ Highlights

  • Remote SSH compute hosts. Register a remote machine over SSH as a notebook execution target and route runs to it alongside local kernels — the first step of Phase 4 remote compute. (#326)
  • Headless task SDK and CLI. A zero-dependency Node.js package (open-science) exposes a task-automation SDK and headless CLI that share the same daemon, projects, sessions, credentials, and permissions as the desktop app — scriptable workflows without leaving the workbench. (#316)
  • Desktop task notifications. Long-running notebook runs and agent tasks now fire OS-level desktop notifications on completion or failure, so the window doesn't need to stay in the foreground. (#300)
  • Session .ipynb export. Export any notebook session as a standards-compliant .ipynb file — per-tab or download-all — validated against the nbformat spec. (#296, #308, #321)
  • Volcengine Ark provider. ByteDance's Volcengine Ark joins the built-in vendor list. (#289)
  • Reasoning-effort control. Choose the reasoning effort (low / medium / high) per model in Settings, for providers that support it. (#263)

🚀 New Features

  • Remote SSH compute host support — register, connect, and route notebook runs to a remote machine over SSH; SSH aliases are passed correctly as the connection target. (#326, #332)
  • Headless task SDK and CLI — zero-dependency open-science Node.js package: task creation/polling, artifact fetch, output formats, and exit codes reusing the local daemon and permissions. (#316)
  • Desktop notifications — OS-level notifications for task completion and failure, with a notification permission request on first use. (#300)
  • Session export as .ipynb — per-tab notebook export plus a download-all, validated against nbformat spec; hardened against edge cases. (#296, #308, #321)
  • Volcengine Ark vendor — ByteDance Volcengine Ark as an official built-in provider. (#289)
  • MiniMax OpenAI and Responses endpoints — both /v1/chat/completions and /v1/responses endpoints registered for MiniMax. (#290)
  • Reasoning-effort selection — per-model reasoning effort control in Model Settings. (#263)
  • Web project/session deep links — shareable deep links for projects and sessions in the localhost web UI, with external lifecycle sync. (#318, #317)
  • Updater download progress — the in-place updater now shows download size and bytes transferred. (#323)
  • Copy and edit user messages — copy or inline-edit any user message in the conversation. (#274)
  • WebSearch re-enabled for ACP agent — the built-in WebSearch tool is back on for the agent (had been removed in a prior settings change). (#307)
  • Windows CRAN R detection — the notebook runtime now detects a user-installed CRAN R on Windows as a bring-your-own interpreter. (#273)
  • Redesigned agent management UI — the agent/framework management panel in Model Settings is fully redesigned for clarity. (#295)

🔧 Improvements

  • Provider connection test is now proxy-aware (uses net.fetch, same path as agent calls) and framework-aware (validates against the selected backend). (#304, #292)
  • Provider-type dropdown in the add-provider flow opens at the top with a framework-matched default. (#298)
  • Runtime install state is isolated per runtime instance, preventing phantom progress indicators during parallel installs. (#285)
  • MiniMax native-Responses providers drive the Responses endpoint directly instead of going through the bridge. (#291)
  • Agent error display shows the actual error message instead of a generic "Agent run failed". (#272)
  • Bioconda mirror availability is validated before provisioning, avoiding silent failures on connectivity-limited systems. (#306)
  • Claude MCP tool names render correctly in the agent-framework tool-activity cards. (#311)
  • Codex commands appear in the permission-approval UI. (#315)
  • Codex permission grants are kept session-scoped rather than leaking across sessions. (#320)
  • Codex MCP permission identity is restored after an upgrade regression. (#305)
  • Codex native-Responses auth detection (api-key / gateway / subscription) is hardened. (#268)
  • AI-review CI workflow now uses type and ready-to-merge labels instead of the old ai-reviewed. (#269)
  • Node 22 LTS pinned via .nvmrc to avoid silent extract-zip hang on Node 26. (#325)
  • Remote SSH compute review items addressed — race condition, notify wiring, and poll-budget handling. (#331)
  • Scoped permission labels in the workspace UI are now word-wrapped correctly. (22ff744)
  • Cross-framework and OpenCode session resume failures are recovered gracefully. (#299)
  • Windows MAX_PATH budget is reserved for the longest default environment name, fixing provisioning overflow. (#330)

🐛 Bug Fixes

  • Artifact source paths — relative and bare paths passed to write_artifact_file now resolve against the session workspace directory (fixes notebook plots failing to write). (#270)
  • Notebook runtime lifecycle — provisioning, crash recovery, and reset lifecycle hardened; recovery is awaited before a bound runtime is used. (#256, 7dfd72a)
  • Codex subscription model seeding — the isolated subscription model is correctly seeded into CODEX_CONFIG on provider creation. (#301)
  • Windows updater — the in-place updater recovers from old-uninstaller failure during Windows updates. (#275)
  • Reviewer identity — the reviewer now correctly recognises the claude-code reviewer MCP tool identity so it doesn't misclassify its own verification tool calls. (#333)
  • Codex policy-amendment permission boundary — the ACP policy-amendment permission boundary is restored for Codex sessions. (#328)
  • SSH connection target — the SSH alias (not the resolved IP) is now passed as the connection target, fixing reconnect on hosts with dynamic IPs. (#332)

📦 Install

Requirements: macOS 12+ (Apple Silicon or Intel), Linux x64, or Windows 10/11 x64. On first run the app's onboarding wizard checks your environment and can install and configure a Claude-compatible model backend for you. Once installed, the app updates itself in place.

Download the build for your platform from the Assets below:

Platform File
macOS (Apple Silicon) open-science-0.6.0-mac-arm64.dmg
macOS (Intel) open-science-0.6.0-mac-x64.dmg
Linux open-science-0.6.0-linux-x86_64.AppImage / .deb
Windows open-science-0.6.0-win-x64-setup.exe

macOS — first launch. Release builds are Developer ID signed and notarized by Apple, so they open like any other app. (A .app you build yourself is not notarized — see README → Building From Source.)

Windows — first launch (unsigned build). No Authenticode certificate yet, so SmartScreen shows a bypassable "unrecognized app" prompt (More info → Run anyway).

Build from source instead:

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

🧭 What's in this release (maturity)

Aligned with the PRD and Roadmap. This is a minor release advancing Phase 4 remote compute (SSH hosts + desktop notifications) and the Phase 1 CLI/SDK surface.

  • ✅ Implemented: desktop shell (Electron + React + TS) with a tray/single-instance lifecycle, reachable as a desktop app, an optional localhost web UI (with project/session deep links and lifecycle sync), a headless backend mode, and an open-science CLI daemon plus a zero-dependency Node.js task SDK; ACP agent runtime on a pluggable framework backend (Claude Code default, with OpenCode and Codex as alternatives — Codex usable via an existing ChatGPT/Codex subscription login as well as an API key) whose app-managed runtimes are installable, switchable, and uninstallable from Settings; an opt-in reviewer with a bounded fix loop; a persistent multi-kernel notebook runtime (warm Python/R/REPL kernels, app-managed conda environments with offline provisioning and bring-your-own interpreters, crash-recoverable run history, session .ipynb export); project + session persistence with a paginated, session-grouped Files tab; artifacts MCP server; file renderers + notebook/session-notebook viewers + a full-screen preview mode + offline molecule viewer; multi-provider config (built-in vendors Anthropic (Claude), DeepSeek, Zhipu AI (GLM), GLM Coding Plan, Kimi (Moonshot), Kimi For Coding, MiniMax, StepFun, Xiaomi MIMO, SenseNova, Volcengine Ark, plus the OpenRouter aggregation gateway, custom gateways, and local Claude) with per-model image-input metadata and per-model reasoning-effort control; desktop task notifications; remote SSH compute hosts as notebook execution targets; signed-and-notarized macOS packaging plus Windows/Linux, with in-place auto-update and a configurable data-storage location; file-based agent skills; 24 built-in connectors plus custom MCP servers; a guided onboarding wizard; and per-conversation approval profiles.
  • 🚧 Partial: R remains managed-only and environment snapshotting/export is not built; the CLI/SDK surface is a first entry point and task-automation layer, not yet a fully scriptable/embeddable orchestration SDK; provider choice is still gated on the selected framework's endpoint compatibility rather than one unified gateway; remote compute covers SSH hosts for notebook runs but not Slurm/cloud-GPU submission or sub-agent fan-out; skills are local and file-based (no shared/public commons yet); the reviewer verifies a turn's own record but does not yet apply domain-specific citation/units/statistics rubrics.
  • 🗺️ Roadmap (not in this build): public skills commons and savable specialist roles (rest of Phase 3); a fully model-agnostic gateway with unified per-agent routing and a full CLI/SDK entry point (rest of Phase 1); artifact versioning + provenance chain, environment snapshotting/export, and specialist sub-agents (rest of Phase 2); Slurm/cloud-GPU remote compute, sub-agent fan-out, and the rest of the security stack — network sandbox, credential vault, scoped permission tiers (rest of Phase 4).

🐢 Known Limitations

  • R is managed-only. A bring-your-own R interpreter path is not built yet.
  • No environment snapshotting or export yet. Phase 2 work.
  • Remote compute is SSH-only. Slurm, cloud GPU, and sub-agent fan-out are not built yet.
  • Provider choice is per backend, not one unified gateway. Claude Code reaches Anthropic-compatible endpoints; OpenCode and Codex add OpenAI-compatible / Responses ones.
  • The task SDK is a first-generation surface. Task creation, polling, and artifact fetch work; a broader programmatic orchestration API is still open Phase 1 work.
  • Switching backends can't carry a live session over. Open conversations replay to the new backend, but in-flight tool state is not migrated.
  • Skills are local only. No shared/public commons, cross-machine forking, or user-facing version pinning yet.
  • The reviewer is opt-in and record-scoped.
  • Windows builds are unsigned. SmartScreen prompt on first launch. (macOS is notarized.)
  • No provenance chain / artifact versioning yet. Phase 2.
  • No local GPU compute backend.
  • No multi-user / real-time collaboration.

🙏 Acknowledgements

Thanks to @ewen-poch, @imjszhang (session .ipynb export, project/session deep links, and external lifecycle sync), @wen2zhou (remote SSH compute host support), @roxi3906 (redesigned agent management UI), @GiaoLee, and everyone in Discord, X, and Discussions.


What's Changed

  • fix(notebook): harden runtime provisioning, crash recovery, and reset lifecycle by @ewen-poch in #256
  • fix(codex): trust api-key/gateway auth, bound status reads, and gate discarded sign-ins by @ewen-poch in #268
  • fix(notebook): await recovery before bound runtime execution by @ewen-poch in #257
  • fix(artifacts): resolve relative artifact source paths against the session workspace by @ewen-poch in #270
  • test(coverage): expand boundary coverage by @ewen-poch in #271
  • feat(settings): add reasoning effort selection to model settings by @ewen-poch in #263
  • ci(ai-review): replace ai-reviewed with type and ready-to-merge labels by @ewen-poch in #269
  • feat(notebook): detect user-installed CRAN R on Windows by @ewen-poch in #273
  • fix(acp): display actual error message instead of generic 'Agent run failed' by @ewen-poch in #272
  • fix(update): recover from old-uninstaller failure during Windows updates by @ewen-poch in #275
  • feat(providers): add MiniMax OpenAI and Responses endpoints by @ewen-poch in #290
  • feat(providers): add Volcengine Ark official vendor by @ewen-poch in #289
  • feat(workspace): add copy and edit actions to user messages by @ewen-poch in #274
  • fix(codex): drive native-Responses providers directly, not via the bridge by @ewen-poch in #291
  • fix(settings): isolate install state per runtime to prevent phantom progress (#278) by @ewen-poch in #285
  • fix(settings): make provider connection test framework-aware by @ewen-poch in #292
  • fix(acp): recover from cross-framework and opencode session resume failures by @ewen-poch in #299
  • fix(codex): seed isolated subscription model into CODEX_CONFIG by @ewen-poch in #301
  • feat(notebook): export sessions as ipynb by @imjszhang in #296
  • fix(settings): open the provider-type dropdown at the top with a framework-matched default by @ewen-poch in #298
  • docs: update README for 0.5.1 by @GiaoLee in #303
  • fix(notebook): validate bioconda mirror availability by @ewen-poch in #306
  • feat(settings): redesign agent management UI in model settings by @roxi3906 in #295
  • fix(settings): probe provider connection over proxy-aware net.fetch by @ewen-poch in #304
  • feat(settings): re-enable the built-in WebSearch tool for the ACP agent by @ewen-poch in #307
  • fix(agent-framework): render Claude MCP tool names by @ewen-poch in #311
  • fix(notebook): harden the ipynb export against the nbformat spec by @ewen-poch in #308
  • fix(acp): restore Codex MCP permission identity by @ewen-poch in #305
  • fix(workspace): wrap scoped permission labels by @ewen-poch in #313
  • fix(acp): show Codex commands in permission approvals by @ewen-poch in #315
  • fix(acp): keep Codex grants session-scoped by @ewen-poch in #320
  • chore(env): pin Node 22 LTS via .nvmrc by @ewen-poch in #325
  • feat(notifications): add desktop notifications for task completion and failure by @ewen-poch in #300
  • feat(compute): add remote SSH compute host support by @wen2zhou in #326
  • feat(updater): show download size and transferred bytes during update by @ewen-poch in #323
  • fix(storage): reserve Windows path budget for the longest default env by @ewen-poch in #330
  • feat(cli): add headless task SDK and CLI by @ewen-poch in #316
  • feat(web): add project and session deep links by @imjszhang in #318
  • fix(reviewer): recognize claude-code reviewer MCP tool identity by @ewen-poch in #333
  • fix(compute): address remote SSH compute review items by @ewen-poch in #331
  • fix(acp): restore Codex policy-amendment permission boundary by @ewen-poch in #328
  • fix(compute): pass ssh alias (not resolved IP) as connection target by @ewen-poch in #332
  • feat(notebook): tab-scoped ipynb export with download-all by @ewen-poch in #321
  • feat(web): sync external project and session lifecycle by @imjszhang in #317
  • fix(notebook): shorten Windows default runtime prefixes by @ewen-poch in #335
  • chore(release): bump version to 0.6.0 by @ewen-poch in #339

Full Changelog: v0.5.1...v0.6.0