Skip to content

v0.4.0

Choose a tag to compare

@github-actions github-actions released this 18 Jul 12:55
· 3583 commits to main since this release
8bcc7e3

Open Science v0.4.0

This release is about agent portability and trust. You can now run the same workbench on a choice of agent backend — Claude Code or OpenCode — and switch on an opt-in reviewer that audits a finished turn against its own transcript, logs, and artifacts before you rely on it. It also adds batch skill import and a system-tray, single-instance app lifecycle, plus a round of skills-import and data-root hardening. All self-hosted, on your own machine.

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 portability-and-trust release: it makes the underlying agent implementation swappable, adds a specialist reviewer that verifies the agent's own claims, and hardens the skills-import and data-root paths shipped in earlier releases. The public skills commons and a fully model-agnostic gateway remain ahead on the Roadmap.

✨ Highlights

  • Choose your agent backend (Claude Code ↔ OpenCode). The ACP runtime is now generic and delegates every framework-coupled decision — spawn, model config, session setup, permission mapping, MCP transport, endpoint compatibility — to a swappable backend. Claude Code stays the default; OpenCode is selectable under Settings → Agent framework as a fully wired alternative that runs in app-owned, isolated config so your own OpenCode install is never touched, with every side-effecting tool still routed through the app's permission gate. Switching backends starts a fresh agent session (open conversations keep their messages and are replayed to the new backend). Because OpenCode also speaks OpenAI-compatible endpoints, provider and model choices are validated against the selected backend. (#185)
  • An opt-in reviewer that checks the agent's work. After a turn completes, an optional reviewer runs in a clean, tool-restricted context and traces the agent's claims against the real transcript, execution log, and artifacts — reporting structured pass/warn/fail findings instead of recomputing — then can run a bounded, user-abortable fix loop (up to three rounds) that asks the agent to correct flagged findings and re-reviews only the new work. Auto-review is off by default; enable it per session with the composer's Auto-review toggle. (#159, #183)
  • Stays out of the way, and cleans up after itself. A system-tray presence with single-instance locking keeps exactly one running copy, and quitting now awaits cleanup of the agent child process so it can't outlive the app. (#179, #177)
  • Import skills in bulk. Bring in multiple skills at once from several files or a single mega-zip, instead of one package at a time. (#164)

🚀 New Features

  • Pluggable agent-framework backend — a swappable AgentFramework interface behind the ACP runtime, with Claude Code (default) and OpenCode both fully wired: executable detection/install, app-owned XDG-isolated config and auth, a generated per-backend config that injects the app's provider/model, native skill materialization, and app-enforced permission profiles routed through the permission broker. (#185)
  • Specialist reviewer with a fix loop — a clean-context, tool-restricted audit session that reads turn data through a host SDK, applies a "trace, don't recompute" rubric, and emits pass/warn/fail findings via a single structured call, surfaced as a reviewer card and a session reviewer page. (#159)
  • Tray residency + single-instance lock — a system-tray icon, a single-instance guard, and awaited teardown of agent processes on quit. (#179)
  • Batch skill import — select multiple skill files or one mega-zip and import them together in a single pass. (#164)

🔧 Improvements

  • ACP resume failures are now classified into specific, actionable messages instead of one opaque error, and the history preamble is replayed when an interrupted session resumes so the backend regains full context. (#166, #191)
  • OpenCode hardening: the API key is referenced through an environment variable with the app's permission policy enforced above project config, and the managed OpenCode binary is verified runnable before its path is persisted. (#187, #189)
  • Windows runtime detection rejects an unspawnable claude.cmd and prefers a healthy recorded runtime. (#167)
  • In-app "What's new" notes are shown from the CDN manifest, and the condensed notes are injected into the electron-updater feeds so already-installed clients see them too. (#162, #163)
  • External links open regardless of referrer, and the external-link dialog's primary button matches the settings link color. (#180, #165)

🐛 Bug Fixes

  • Cmd/Ctrl+W now closes the active preview tab, then the preview panel, before closing the window. (#182, #184)
  • Skills-import hardening: reject path traversal in custom-connector and zip imports, cap resource use for zip and GitHub imports, parse SKILL.md frontmatter authored with CRLF line endings, and escape personal-skill frontmatter with a source-labelled detail badge. (#170, #171, #172, #181)
  • Data-root storage: migration is hardened against partial/stale copies, and reviewer artifacts plus an absolute dataRoot now resolve correctly on a configured data root. (#173, #174)
  • An adopted agent session is now closed by its own id on delete, so removing it can't tear down the wrong session. (#188)

📦 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-0.4.0-mac-arm64.dmg
macOS (Intel) open-science-0.4.0-mac-x64.dmg
Linux open-science-0.4.0-linux-x86_64.AppImage / .deb
Windows open-science-0.4.0-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:

  • ✅ Implemented: desktop shell (Electron + React + TS) with a tray/single-instance lifecycle; ACP agent runtime on a pluggable framework backend (Claude Code default, OpenCode alternative) with typed tool-activity rows and a permission gate; an opt-in reviewer with a bounded fix loop; one persistent Python kernel with replayable run history; project + session persistence; artifacts MCP server; file renderers + notebook/session-notebook viewers + offline molecule viewer; multi-provider config (built-in vendors Claude, DeepSeek, Zhipu AI (GLM), MiniMax, Kimi (Moonshot), plus custom gateways and local Claude) over Anthropic-compatible endpoints on any backend and OpenAI-compatible endpoints via OpenCode; signed-and-notarized macOS packaging plus Windows/Linux, with in-place auto-update and a configurable data-storage location; file-based agent skills with a / selector and batch import; 24 built-in connectors (23 upstream-aligned life-science connectors at 200+ tools + the molecule viewer) plus custom MCP servers; a guided first-run onboarding wizard; and per-conversation approval profiles.
  • 🚧 Partial: provider choice is now split across two backends (Claude Code speaks Anthropic; OpenCode adds OpenAI-compatible) but is still gated on the selected framework's endpoint compatibility rather than one unified gateway; skills are local and file-based (no shared/public commons, cross-machine forking, or version pinning 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 CLI/SDK entry point (rest of Phase 1); artifact versioning + provenance chain, additional kernels (R), and specialist sub-agents (Phase 2); remote compute and the rest of the security stack — network sandbox, credential vault, scoped permission tiers (Phase 4).

🐢 Known Limitations

  • Provider choice is per backend, not one unified gateway. Claude Code reaches Anthropic-compatible endpoints; OpenCode adds OpenAI-compatible ones. The fully model-agnostic gateway with unified per-agent routing is still open Phase 1 work.
  • Switching backends can't carry a live session over. Open conversations keep their messages and get replayed to the new backend, but in-flight tool state is not migrated — switching starts a fresh agent session.
  • Skills are local only. You can create, edit, and import them on one machine, but there is no shared/public commons, no cross-machine forking, and no user-facing version pinning yet.
  • The reviewer is opt-in and record-scoped. It traces a turn's own claims against its transcript/logs/artifacts; it is off by default and does not yet run domain-specific citation/units/statistics checks.
  • Windows builds are unsigned. Windows has no Authenticode cert yet, so it shows a bypassable SmartScreen prompt on first launch. (macOS is notarized.)
  • No provenance chain / artifact versioning yet. Lands in Phase 2.
  • No local compute backend. The bundled biomodel skills (AlphaFold2, ESMFold2, DiffDock, …) ship as reference/instruction docs; this build can't execute their GPU/model tooling locally.
  • No multi-user / real-time collaboration. Sharing is via export/import.

🙏 Acknowledgements

Thanks to @ewen-poch and @wen2zhou, and to everyone in Discord, X, and Discussions.


What's Changed

  • fix(update): show in-app release notes from the CDN manifest on the electron-updater path by @ewen-poch in #162
  • feat(reviewer): add specialist auto-review with fix loop by @wen2zhou in #159
  • fix(update): inject release notes into electron-updater feeds so installed clients show them by @ewen-poch in #163
  • feat(skills): batch-import multiple skills from files or a mega-zip by @ewen-poch in #164
  • fix(ui): match external-link dialog primary button to settings link color by @ewen-poch in #165
  • fix(acp): classify agent session resume failures by @ewen-poch in #166
  • fix(win): reject unspawnable claude.cmd in detection and prefer healthy recorded runtime by @ewen-poch in #167
  • ci: add stage-runtime-bundle workflow (dispatch-only) by @ewen-poch in #175
  • fix(acp): kill agent child on app quit so it cannot outlive the app by @ewen-poch in #177
  • fix(storage): reviewer artifacts on data root + absolute dataRoot by @ewen-poch in #174
  • fix(storage): harden data-root migration against partial/stale copies by @ewen-poch in #173
  • fix(main): open external links regardless of referrer by @ewen-poch in #180
  • fix(skills): escape personal-skill frontmatter and label detail badge by source by @ewen-poch in #172
  • fix(skills): cap resource use for zip and GitHub skill imports by @ewen-poch in #171
  • fix(skills): parse SKILL.md frontmatter authored with CRLF line endings by @ewen-poch in #181
  • fix(skills): harden custom-connector and zip import paths against traversal by @ewen-poch in #170
  • fix(reviewer): default auto-review to off (opt-in) by @ewen-poch in #183
  • fix(window): Cmd/Ctrl+W closes the preview panel before the window by @ewen-poch in #182
  • fix(window): Cmd/Ctrl+W closes the active preview tab before the panel by @ewen-poch in #184
  • feat: pluggable agent framework (Claude Code ↔ OpenCode) by @ewen-poch in #185
  • feat(lifecycle): add tray residency, single-instance lock, and awaited quit cleanup by @ewen-poch in #179
  • fix(agent-framework): secure OpenCode API key and enforce permission policy by @ewen-poch in #187
  • fix(acp): close adopted agent session by its own id on delete by @ewen-poch in #188
  • fix(settings): verify managed OpenCode binary runs after install by @ewen-poch in #189
  • test(agent-framework): cover reviewer promptPrefix and preload channels by @ewen-poch in #190
  • fix(acp): replay history preamble on interrupted-session resume by @ewen-poch in #191
  • chore: bump version to 0.4.0 by @ewen-poch in #186

New Contributors

Full Changelog: v0.3.1...v0.4.0