v0.18.1

Open Science v0.18.1

Session-scoped compute host discovery and execution-target selection with CLI support, a redesigned error-notice component at app startup, improved message-center alerts with project and task context, collapsed long user messages, empty initial review checks, French localization, an imported-skills group menu, and CLI commands for application updates and project agent context — alongside faster settings and startup through panel caching and runtime-probe reuse, concurrent message and branch reconciliation, connector artifact provenance, hardened local HTTP boundaries, and updater recovery guidance.

Open Science is an open-source, local-first AI research workbench for scientists and researchers. It enables reproducible, inspectable research across models with scientific AI agents, Python and R execution, scientific data connectors, and cross-platform support for macOS, Windows, and Linux.

v0.18.1 is a patch release. Compute hosts become session-scoped: each session now holds separate enabled (agent-discoverable) and selected (execution-target) host sets, with a compact composer popover for target selection, CLI and Task API support for choosing configured hosts when creating or continuing a session, and trusted ownership enforced across every host operation. The startup failure page is redesigned as a reusable, data-driven error-notice component applied to the database startup gate. Message-center alerts gain project, task, and preview context, group unread items first with a live toast anchored to the visible bell, and keep item actions independent of read writes. Long user messages collapse after twelve rendered lines with an accessible show-more control. The reviewer can now return an empty check set on initial review when a turn contains nothing checkable, instead of manufacturing a passing check. French joins the interface languages with a complete catalog, native messages, and matching project guides. The CLI gains update for terminal-driven application updates and project update / --agent-context inputs for managing project agent context headlessly. On the performance side, settings panels are cached and lazy-loaded, runtime probes are reused across startup, and the application critical path is shortened. Persistence reconciles concurrent message and branch changes, connector-written files are recorded in artifact provenance, local HTTP boundaries are hardened, and blocked installs get explicit recovery guidance.

✨ Highlights

• Session-scoped compute hosts. Each session keeps separate enabled and selected compute-host sets — enabled hosts are discoverable by the agent and labeled selected or available; the composer gains a compact execution-target popover; the CLI and Task API accept --compute-host selections when creating or continuing a session; every host command, download, submission, status, and result operation enforces trusted project/session/provider ownership. Legacy sessions migrate their enabled hosts as selected. (#1496)
• Redesigned error notices. Startup failures now render through a reusable, data-driven error-notice component — a fixed brand mark, semantic tone palettes, optional title/description/error-code sections — applied first to the database startup gate and reusable by any future error surface. (#1477)
• Message-center alerts with context. Notification rows show project, task/session title, and a one-line event preview; unread items group first while preserving newest-first order; a six-second live toast anchors to the currently visible bell, re-anchors after navigation, and aggregates bursts; deleted sessions show stable unavailable context. (#1491, #1471)
• French localization. A complete French catalog with native plural-category coverage (one/many/other), host-language detection, persisted selection, native main-process messages, and date formatting joins Chinese (Simplified and Traditional), Japanese, and Korean, with matching translated project guides. (#1433, #1478)

🚀 New Features

• Session-scoped compute host discovery and selection — separate per-session enabled and selected host sets with legacy migration; agent-visible hosts labeled selected/available; composer controls and a compact execution-target popover with per-host settings links; CLI --compute-host (repeatable) and Task API/SDK selection of already-configured hosts; host creation, credentials, probing, editing, and deletion stay desktop-only. (#1496)
• Redesigned error notice — a reusable data-driven component with brand mark, semantic tones, and optional sections, applied to the database startup gate. (#1477)
• Improved message center — project, task, and preview context on message rows; unread-first grouping with chronological return after read; a six-second live toast anchored to the visible bell with hover/focus pause and burst aggregation; notification failures isolated from the app. (#1491)
• Empty initial review checks — the reviewer may finish an initial assessment with zero checks when a turn has nothing checkable (greetings, acknowledgments, clarifications), while tracked rechecks still require the exact tracked set. (#1495)
• French localization — complete renderer catalog with guards for plural categories, terminology, and safety copy; fr registered across locale detection, native messages, formatting, and persisted selection. (#1433)
• Imported skills group menu — a neutral Import dropdown on the Imported group header shares upload, GitHub, and installed-folder actions with the Add skill menu. (#1488)
• Collapsed long user messages — user messages collapse after 12 rendered lines with an accessible Show more/Less control, recompute on width change, attachments kept outside the collapsed region, and structured prompts collapsed to a plain-text summary. (#1479)
• CLI application update — open-science update (and --json) drives the existing check/download/verify/apply workflow from the terminal. (#1487)
• CLI project agent context — project create --agent-context[--file] and a new project update command for name, description, agent-context replacement, and explicit clearing, carried through the SDK and HTTP Task API with optimistic concurrency; public responses expose hasAgentContext without contents. (#1464)

🔧 Improvements

• Settings panels are cached and lazy-loaded, cutting settings navigation cost. (#1492)
• Runtime probes are reused across startup with session-recovery tracing, shortening first-launch time. (#1469, #1460)
• Windows E2E checks run faster. (#1458)
• README gains improved SEO and AI citability structure. (#1493)

🐛 Bug Fixes

• Connector artifact provenance recorded. Files written through writable connector filesystems are now published as artifacts with recorded identities, so provenance covers connector-produced outputs. (#1490, 80eb7b9)
• Settings facade contract kept. Settings load actions stay on the facade contract. (#1494)
• Blocked installs recoverable. The updater guides recovery when a pending install blocks a new one. (#1481)
• Revision conflicts transported as outcomes. Session revision conflicts surface as typed outcomes instead of errors. (#1480)
• Concurrent message and branch changes reconciled. Concurrent edits to a session's messages and branches merge disjoint changes instead of clobbering. (#1474, #1470)
• Tool layout shift prevented. Streaming tool cards no longer shift layout on arrival. (#1476)
• Versioned JSON reads classified. (#1475)
• Overlay preload capabilities isolated. The window-find overlay's preload surface is isolated to its needs. (#1472)
• Local HTTP boundaries hardened. (#1466)
• Web SDK lifecycles separated. Transport and operation lifecycles no longer tangle. (#1465)
• Claude 4xx reports hidden. Provider 4xx noise no longer surfaces as reports. (#1467)
• Durable JSON publication unified / legacy move prompt failures handled / staged migration recovery expected. (#1452, #1463, #1473)
• Artifact downloads published atomically. (#1461)
• Deterministic provider errors stop retrying. (#1459)
• Stop state scoped to session. (#1462)
• Stale edits can no longer create resources. (#1451)
• Non-ASCII micromamba caches avoided. Notebook package caches avoid non-ASCII paths that break some toolchains. (#1483)
• Collapsed message preview aligned. (#1482)

📦 Install

Requirements: macOS 12+ (Apple Silicon or Intel), Linux x64, or Windows 10/11 x64. On first run, the onboarding wizard checks the environment and can install and configure an app-managed agent runtime. Once installed, the app can update itself in place.

Download the appropriate package from the Assets section below:

Platform	Package
macOS (Apple Silicon)	DMG for ARM64
macOS (Intel)	DMG for x64
Linux	AppImage or Debian package for x64
Windows	Installer for x64

macOS — first launch. Official release builds are Developer ID signed and notarized by Apple, so they open like other trusted applications. A locally built copy is not notarized and may require approval through macOS Privacy & Security.

Windows — first launch (unsigned build). No Authenticode certificate yet, so SmartScreen shows a bypassable "unrecognized app" prompt (More info → Run anyway). Verify that the package came from the official release page before continuing.

Build from source instead:

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

🧭 What's in this release (maturity)

• ✅ Implemented: a local-first desktop, localhost-web, headless, CLI, and task-SDK surface over persistent projects and sessions with selectable message branches, branching into a new session from user messages or completed agent messages with persisted source lineage, reversible archiving, project pinning, collapsed long user messages, and persistent side conversations; selectable Claude Code, OpenCode, and Codex agent frameworks behind a shared provider turn-adapter interface; production subagent delegation with durable messaging, restart recovery, structured output, artifact and review evidence, and camelCase delegation APIs; review-gated session plans with CLI plan controls; a unified composer lane with a session-scoped message queue and active-specialist emphasis; hot-switching of compatible models and providers; multi-provider model configuration with a dedicated Vision model selector and persistent evidence relay; a configurable reviewer model policy with an isolated review runtime, durable assessment snapshots, and empty initial checks; a token usage dashboard; context-window composition insights; persistent Python/R/REPL kernels with bounded run-history payloads, separated approval and execution state, session-scoped remote SSH execution with key or password authentication, harvest ceilings, and a host viewImage capability; frame-scoped Host JavaScript artifact queries, session diagnostics, and model introspection with camelCase Host LLM results; immutable artifact versions with artifact lineage access and connector artifact provenance; multi-format previews with view-in-context navigation and plan-artifact rendering; file-based skills with conversational creation, save-as-skill turns, direct user folder support, a redesigned management experience, an imported group menu, and confirmed bulk deletion; cross-resource catalog tags with favorites and drag ordering; permissioned scientific connectors with immutable invocation identities, generated local IDs, and aligned catalog interactions; opt-in review and bounded correction; personal specialist agent profiles with a scoped signed marketplace, instant cached browsing, capability icons, quick appearance editing, and capability-row navigation; scoped permission management; local folder access with cross-drive browsing; Chinese (Simplified and Traditional), Japanese, Korean, and French interfaces with multilingual README and contribution guides; global proxy settings with connectivity diagnostics; prominent update reminders with CLI-driven updates; a run-marks navigation rail; message-center alerts with project and task context; notifications with attention reasons; empty-conversation placeholder banners; a redesigned startup error notice; contextual GitHub star prompts; conversation export as Markdown and PDF; a project-scoped command palette; CLI Codex device login and project agent-context management; smooth live response rendering; collapsible side panels; split-view file preview; desktop notifications; a cross-surface notification message center; structured agent clarification cards; live session status; and session keyboard shortcuts.
• 🚧 Partial: R remains managed-only; provider choice remains constrained by the active framework's endpoint compatibility; remote compute is SSH-only; skills remain local (no hosted public discovery commons); and review is opt-in and record-scoped.
• 🗺️ Roadmap: a unified model gateway, a hosted public skills and specialist discovery commons, Slurm/cloud-GPU execution, stronger sandboxing and credential isolation, and collaborative research workflows.

🐢 Known Limitations

• R is managed-only. A bring-your-own R interpreter path is not built yet.
• Remote compute is SSH-only. Slurm and cloud GPU submission are not built yet.
• Provider choice is per framework, not one unified gateway. The available protocol depends on the selected agent backend.
• Hot-switching applies only to registered compatible targets. Framework, auth-lane, wire-route, or unsafe capability changes still require a reconnect.
• Code reconstruction is LLM-generated. It does not replace deterministic reproduction; portable environment locks and full-fidelity session replay remain open.
• No hosted public specialist discovery commons. Specialist packages are portable across machines via import/export and the signed marketplace; what is not built yet is a hosted public discovery and forking hub.
• The task SDK is a first-generation surface. Task creation, polling, artifact retrieval, run progress, and cancellation work; broader orchestration remains open.
• Switching agent backends cannot transfer in-flight tool state. Existing conversation history can replay, but a running action is not migrated.
• Skills are local only. There is no shared public commons, cross-machine forking, or user-facing version pinning yet.
• The reviewer is opt-in and record-scoped. It does not replace domain-specific validation of citations, units, statistics, or methods.
• Scoped permissions cover allow-grants only. Network sandboxing, directory-level file access control, and a credential vault are not built yet.
• Windows builds are unsigned. SmartScreen may warn on first launch; official macOS builds are notarized.
• No local GPU compute backend.
• No multi-user real-time collaboration.

🙏 Acknowledgements

Thanks to @ewen-poch, @wen2zhou, @roxi3906, @daanveer-tech, and everyone in Discord, X, and Discussions.