v0.5.1

Open Science v0.5.1

A stabilization release on top of v0.5.0's compute-and-backend work: the Codex backend now signs in with an existing ChatGPT/Codex subscription (no OpenAI API key required), the model layer gains SenseNova and the GLM Coding Plan endpoint, the Files tab gets indexed pagination and session grouping, and a broad round of Windows reliability and lifecycle hardening lands across the notebook runtime, Codex managed install, sessions, and the close/quit flow. 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 patch release: no breaking changes, no new maturity claims beyond v0.5.0. It closes the "Codex is API-key-only" gap called out in the v0.5.0 known limitations, broadens providers, and hardens the paths that shipped in v0.5.0. A fully model-agnostic gateway, artifact provenance, and the public skills commons remain ahead on the Roadmap.

✨ Highlights

• Codex signs in with a subscription, not just an API key. Add a Codex provider backed by the locally installed Codex CLI and reuse an existing ChatGPT/Codex subscription login — either your existing Codex profile or a separate isolated login — instead of configuring an OpenAI API key. Authentication homes and provider identities stay separated, and Codex provider entries can be edited and deleted like any other. (#227, #264)
• Two more model vendors. SenseNova (SenseTime, China region — Messages + Chat Completions endpoints, sensenova-6.7-flash-lite multimodal) joins the built-in vendor list, and the GLM Coding Plan endpoint lands as its own entry under Zhipu (GLM). (#259, #246)
• A navigable Files tab. The project file library gets indexed pagination and session grouping, so large projects list and scroll their uploads and generated artifacts without loading everything at once. (#217)
• Windows reliability pass. Managed R runtime MAX_PATH overflow, notebook shell execution, managed-Codex install across drives, and new-session creation in managed workspaces are all fixed for Windows. (#239, #252, #236, #228)

🚀 New Features

• Codex subscription authentication — Codex subscription providers backed by the installed Codex CLI and ACP bridge; choose an existing Codex profile or a separate isolated login, with authentication homes and provider identities kept separate, and full edit/delete parity with other providers. (#227)
• SenseNova provider — SenseTime's SenseNova as an official vendor (China region, Anthropic-compatible Messages and OpenAI-compatible Chat Completions on one host), with sensenova-6.7-flash-lite marked multimodal and deepseek-v4-flash text-only. (#259)
• GLM Coding Plan endpoint — a dedicated GLM Coding Plan entry under Zhipu (GLM), across the Global (Z.AI) and China (BigModel) consoles. (#246)
• Files pagination + session grouping — indexed pagination and per-session grouping in the project Files tab. (#217)
• Close/quit confirmation — an explicit confirm step before closing to tray or quitting while work is running, so an in-flight run isn't ended by accident. (#238)
• Unified composer controls — the composer permission mode and the auto-review toggle are merged into one menu. (#237)

🔧 Improvements

• Reviewer hardening: audit isolation and fix-loop integrity are tightened so a review runs cleanly against its own record. (#255)
• The reviewer/agent now recovers from context-overflow conditions that providers surface only as opaque error slugs, instead of failing the turn. (#254)
• claude-agent-acp upgraded to 0.60.0 to handle command_lifecycle events. (#251)
• Codex detection now distinguishes a native Codex install from the adapter in its status messages. (#249)
• The home page hides the account button and the environment-status banner is overlaid rather than clipping the bottom of the workspace. (#240, #250)
• Storage internals rename ActiveSessionInfo.projectName to projectId, resolving display names in the renderer. (#241)
• CI raises the Vitest testTimeout to 15s to remove a flaky-timeout failure. (#258)

🐛 Bug Fixes

• Codex subscription auth — validation and the sign-in flow are hardened after the initial subscription-auth landing. (#264)
• Artifact write paths — write_artifact_file now resolves relative and bare paths against the notebook data directory, fixing notebook-generated plots that failed to write. (#262)
• Windows managed R runtime — confirmed MAX_PATH cache overflow during provisioning is fixed, with Windows R discovery and conda DLL activation scoped correctly. (#239)
• Windows notebook shell — shell execution is supported on Windows. (#252)
• Windows managed Codex install — falls back to copy-and-delete when rename throws EPERM/EXDEV across drives. (#236)
• Managed workspaces — new agent sessions are created inside the managed workspace instead of the wrong root. (#228)
• Session recovery — an unreadable session is preserved on recovery rather than discarded. (#253)
• Create-session diagnostics — bounded, hardened diagnostics are added for agent create-session failures. (#232)

📦 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-0.5.1-mac-arm64.dmg
macOS (Intel)	open-science-0.5.1-mac-x64.dmg
Linux	open-science-0.5.1-linux-x86_64.AppImage / .deb
Windows	open-science-0.5.1-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. Maturity is unchanged from v0.5.0 — this is a patch release. The only change worth noting: the Codex backend now authenticates with an existing ChatGPT/Codex subscription login, so it is no longer API-key-only.

• ✅ Implemented: desktop shell (Electron + React + TS) with a tray/single-instance lifecycle, reachable as a desktop app, an optional localhost web UI, a headless backend mode, and an open-science CLI daemon; ACP agent runtime on a pluggable framework backend (Claude Code default, with OpenCode and Codex as alternatives — Codex now 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); 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, plus the OpenRouter aggregation gateway, custom gateways, and local Claude) with per-model image-input metadata; 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/browser surface is a first entry point, not yet a scriptable/embeddable SDK; provider choice is still gated on the selected framework's endpoint compatibility (Claude Code speaks Anthropic; OpenCode and Codex add OpenAI-compatible / Responses) rather than one unified gateway, and Codex's per-vendor live function-tool verification is an ongoing release gate; 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); remote compute and the rest of the security stack — network sandbox, credential vault, scoped permission tiers (Phase 4).

🐢 Known Limitations

• R is managed-only. You get a warm R kernel against an app-managed environment; a bring-your-own R interpreter path is not built yet.
• No environment snapshotting or export yet. You can create, bind, and remove environments, but capturing a reproducible snapshot/lockfile of one is Phase 2 work.
• Codex live verification is ongoing. Codex now runs against both API-key providers and a ChatGPT/Codex subscription login; per-vendor live function-tool verification remains an ongoing release gate.
• Provider choice is per backend, not one unified gateway. Claude Code reaches Anthropic-compatible endpoints; OpenCode and Codex add OpenAI-compatible / Responses ones. The fully model-agnostic gateway is still open Phase 1 work.
• The web UI is loopback-only, and the CLI is an entry point, not an SDK. Both bind to 127.0.0.1; there is no LAN/remote-access mode, and a scriptable/embeddable interface 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.
• 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. It traces a turn's own claims against its transcript/logs/artifacts; it is off by default and does not yet run domain-specific 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 GPU compute backend. The bundled biomodel skills 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, @bobo-xxx (first contribution — Codex ChatGPT subscription authentication and managed-workspace session creation), @roxi3906, and @GiaoLee, and to everyone in Discord, X, and Discussions.

---

What's Changed

• fix(acp): create new sessions in managed workspaces by @bobo-xxx in #228
• feat: unify composer permission mode and auto-review into one menu by @roxi3906 in #237
• chore: hide account button on home page by @ewen-poch in #240
• fix(managed-codex): fall back to copy+delete when rename throws EPERM or EXDEV on Windows by @ewen-poch in #236
• feat(window): confirm before close-to-tray / quit with running work by @ewen-poch in #238
• refactor(storage): rename ActiveSessionInfo.projectName → projectId; resolve names in the renderer by @ewen-poch in #241
• fix(window): harden close/quit confirmation lifecycle (orphan + hung-modal fallbacks) by @ewen-poch in #243
• feat(providers): add GLM Coding Plan endpoint under Zhipu (GLM) by @ewen-poch in #246
• fix(acp): add diagnostics for agent create-session failures by @ewen-poch in #232
• feat(files): add indexed pagination and session grouping by @roxi3906 in #217
• fix(runtime): Windows managed R runtime MAX_PATH failures by @ewen-poch in #239
• fix(deps): upgrade claude-agent-acp to 0.60.0 to handle command_lifecycle events by @ewen-poch in #251
• fix(workspace): overlay env status banner to stop clipping page bottom by @ewen-poch in #250
• fix(codex): distinguish native Codex from adapter in detection messages by @ewen-poch in #249
• fix(files): preserve unreadable session recovery by @ewen-poch in #253
• fix(acp): recover context overflow surfaced via provider error slugs by @ewen-poch in #254
• fix(notebook): support Windows shell execution by @ewen-poch in #252
• test: raise vitest testTimeout to 15s to fix CI flake by @ewen-poch in #258
• fix(reviewer): harden audit isolation and fix-loop integrity by @GiaoLee in #255
• ci(review): unify automated AI reviews by @ewen-poch in #260
• feat(providers): add SenseNova official vendor by @ewen-poch in #259
• feat(codex): support ChatGPT subscription authentication by @bobo-xxx in #227
• ci(review): satisfy Claude action auth validation by @ewen-poch in #261
• ci(review): grant pull-requests write to AI review label jobs by @ewen-poch in #265
• fix(artifacts): resolve artifact write paths against the notebook data dir by @ewen-poch in #262
• fix(codex): harden Codex subscription auth validation and sign-in flow by @ewen-poch in #264
• docs(contributing): improve pull request guidance by @ewen-poch in #266
• chore(release): bump version to 0.5.1 by @ewen-poch in #267

New Contributors

• @bobo-xxx made their first contribution in #228

Full Changelog: v0.5.0...v0.5.1