Repository navigation
v0.7.3
Open Science v0.7.3
A polish and reach release on top of v0.7.2: native context compaction for long sessions, a dark-mode theme, a whole-window find bar, large-file streaming uploads, a Grok (xAI) provider, and framework-aware custom-gateway defaults.
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 and no Roadmap phase status changes. It adds native context compaction (so long sessions self-recover instead of stalling at the context ceiling), a dark-mode theme with synchronous first-render application, a Cmd/Ctrl+F find bar that searches the entire workspace, streaming uploads up to 10 GB per file, a Grok (xAI) provider, and custom-gateway API-format defaults derived from the active agent framework, plus a set of bug fixes hardening Codex model metadata, conversational skill imports, remote compute sourcing, and provider validation.
✨ Highlights
- Native context compaction. Claude Code and OpenCode sessions now invoke the framework's own compaction mechanism — automatically at 90% context usage, or on demand from the context-indicator popover — so long-running sessions stay within the model's context window without losing the conversation. (#463)
- Dark mode. A warm, low-chroma dark theme is available under Settings → General, applied synchronously before first render (no flash), covering the shell, transcript, and all renderer palettes. (#399)
- Whole-window find.
Cmd/Ctrl+Fopens an overlay search bar over the entire workspace — agent transcripts, notebook output, and rendered pages — with a live match counter, next/previous navigation, and last-query refill. (#466) - Large file uploads. Attachments up to 10 GB are streamed directly into managed storage on desktop (native file path resolution in preload) and via a chunked begin/append/finish protocol on web, replacing the previous 50 MB whole-file base64 cap. (#459)
- Grok (xAI) provider. xAI is now a first-class built-in vendor with curated Grok models, branded icon, and Chat Completions + Responses endpoints at
api.x.ai. (#457)
🚀 New Features
- Native context compaction — a framework-level compaction strategy routes Claude Code and OpenCode sessions through their native
/compactcontrol turn (Codex remains framework-managed); compaction fires automatically when reported context usage reaches 90%, and a hover/focus-accessible context popover exposes a manual Compact context action with progress and cancellation; overflow recovery prefers native compaction and falls back to the existing fresh-session/text-replay path; native control turns stay out of the visible transcript while preserving usage updates and safe per-session lifecycle locks. (#463) - Dark mode — a complete theme system with light (
:root) and dark (.dark) token sets for all shell, transcript, and streamdown/agent-markdown palettes, following a warm, low-chroma aesthetic that preserves the paper-like feel; the preference is stored inlocalStorageand applied synchronously before React mounts (both Electron and web builds) to prevent flash-of-unstyled-content; all hardcoded structural colors (bg-white,border-[...]) were converted to semantic tokens, with a toggle in Settings → General → Appearance. (#399) - Whole-window find bar — a dedicated
WebContentsViewoverlay owned by the main window, so the query text is never part of the page being searched;Cmd/Ctrl+Fopens the bar (intercepted in main'sbefore-input-event, only while a searchable Workspace is mounted and the renderer is responsive); search-as-you-type with a livecurrent / totalmatch counter,Enter/Shift+Enterfor next/previous,Escor theXbutton to close, and reopening refills the last query. (#466) - Large file streaming uploads — raises the managed-storage limit to 10 GB per file (shown before selection); on desktop, native
Filepaths are resolved in preload and streamed directly into managed storage without exposing source paths to the renderer; for Web, clipboard, and pathlessFileobjects, an 8 MB begin/append/status/finish protocol with offset validation, bounded retry, progress, and cancellation is used; legacy whole-file base64 staging is removed; partial uploads stage into exclusive.partfiles with cleanup of abandoned partials; per-file queued/uploading/error/cancelling state and progress are shown in Composer; model payloads stay bounded (small text embedded, large text gets a bounded preview plusresource_link, dataset binaries and video are link-only, images and PDFs above the auto-processing threshold are link-only). (#459) - Grok (xAI) provider — added immediately after Anthropic in the official catalog, with xAI Chat Completions and Responses endpoints at
https://api.x.ai/v1; curatedgrok-4.5,grok-4.3, andgrok-build-0.1with their documented context, reasoning, and image-input capabilities; a bundled branded Grok SVG replaces the generic provider placeholder. (#457) - Framework-aware custom-gateway defaults — the Custom Gateway default API format (Messages, Chat Completions, or Responses) is now derived from the active agent framework's advertised endpoints, applied only to blank or new Onboarding drafts and when Custom Gateway is newly selected; a saved custom provider's API format is preserved when opened for editing, and user-edited Onboarding fields are preserved across remounts and framework changes. (#461)
- Conversation import toggle — skill conversation imports can be toggled per session, giving users explicit control over whether the agent may request a package import from an attachment. (#489)
🔧 Improvements
- Subscription-first AI review — the CI AI-review pipeline now runs a single subscription-authenticated review first, reducing redundant review runs and token consumption on every PR. (#455)
- OpenRouter model catalog refresh — the bundled OpenRouter model catalog is updated with the latest stable models. (#456)
🐛 Bug Fixes
- Codex native Responses model metadata — Codex now registers native Responses model metadata correctly, so models served through the native Responses path show the right instructions and capabilities instead of falling through to a generic profile. (#451)
- Referenced-upload skill imports —
@-referenced uploads can now be used as skill import sources without being rejected as invalid candidates. (#452) - Conversational import routing hardening — the HTTP host and IPC routing for agent-requested skill imports is hardened against malformed and cross-origin requests, so an invalid or spoofed import request cannot reach the import pipeline. (#453)
- Multi-protocol provider validation — custom providers configured for different API protocols (Messages, Chat Completions, Responses) are now validated reliably against the active backend's endpoint compatibility, preventing false validation failures and misconfigured providers from passing the connection test. (#460)
- Codex imported subscription routing — Codex sessions that authenticate via an imported ChatGPT/Codex subscription login now route requests through the correct provider, instead of dropping back to an API-key path. (#462)
- Electron path.txt trailing newline — a trailing newline in
electron'spath.txtcould resolve to an invalid Electron binary path on some platforms; it is now stripped before use. (#464) - Whole-window find overlay hardening — the find overlay no longer intercepts keystrokes meant for the page when the Workspace is not mounted or the renderer is unresponsive, and the overlay geometry is recalculated on window resize and full-screen transitions. (#467)
- Remote compute Bash config sourcing — remote SSH workloads now source the target user's Bash profile before dispatching a notebook job, so conda and other environment activations that depend on shell init scripts are present on the remote host. (#470)
📦 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) | aipoch-open-science-0.7.3-mac-arm64.dmg |
| macOS (Intel) | aipoch-open-science-0.7.3-mac-x64.dmg |
| Linux | aipoch-open-science-0.7.3-linux-x86_64.AppImage / aipoch-open-science_0.7.3_amd64.deb |
| Windows | aipoch-open-science-0.7.3-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 on top of v0.7.2. It adds native context compaction for long sessions, a dark-mode theme, a whole-window find bar, large-file streaming uploads, a Grok (xAI) provider, and framework-aware custom-gateway defaults. No Roadmap phase changes status.
- ✅ Implemented: desktop shell (Electron + React + TS) with a tray/single-instance lifecycle and a dark-mode theme, 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-scienceCLI 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, and Claude via a subscription with both shared (browser OAuth into~/.claude) and isolated (claude setup-token) sign-in) whose app-managed runtimes are installable, switchable, and uninstallable from Settings; tool-activity cards grouped under agent-declared purpose titles with a live context-usage indicator and native context compaction (automatic at 90% or manual); a whole-window find bar; an opt-in reviewer with a bounded fix loop (AI-review pipeline now subscription-first); 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.ipynbexport); project + session persistence with a paginated, session-grouped Files tab and session pinning; artifacts MCP server; file renderers (incl. Office DOCX/XLSX/PPTX in isolated processes, a zoom/pan image viewer, and a crisp, zoomable PDF preview) + notebook/session-notebook viewers + a full-screen preview mode + offline molecule viewer; file-based agent skills (create, edit, package upload, GitHub preview/import, installed-global-skill import with candidate preview, and agent-requested package imports in a session) with explicit/selection; multi-provider config (built-in vendors Anthropic (Claude), Grok (xAI), DeepSeek, Zhipu AI (GLM), GLM Coding Plan, Kimi (Moonshot), Kimi For Coding, MiniMax, StepFun, StepFun Step Plan, Xiaomi MIMO, SenseNova, Volcengine Ark, and the OpenRouter aggregation gateway), plus custom gateways whose default API format is derived from the active framework, and Claude / Codex subscription logins; streaming uploads up to 10 GB per file; remote SSH compute hosts for notebook runs; in-place background auto-update; a configurable data-storage location; a guided first-run onboarding wizard; desktop notifications on task completion/failure. - 🚧 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; native context compaction covers Claude Code and OpenCode but not Codex (which remains framework-managed).
- 🗺️ 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.
- Context compaction is framework-dependent. Claude Code and OpenCode use their native compaction; Codex sessions are framework-managed.
- 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, @roxi3906, @LJLJ111, @imjszhang, @wen2zhou, @GiaoLee, and everyone in Discord, X, and Discussions.
What's Changed
- fix(codex): register native Responses model metadata (#451)
- fix(skills): allow importing referenced uploads (#452)
- fix(skills): harden conversational import routing (#453)
- feat(providers): update OpenRouter model catalog (#456)
- feat(providers): add Grok xAI provider (#457)
- ci(review): run one subscription-first AI review (#455)
- feat(uploads): support large file attachments (#459)
- feat(model-settings): default custom API format by agent framework (#461)
- fix(model-settings): validate multi-protocol providers reliably (#460)
- feat(skills): add conversation import toggle (#489)
- fix(codex): preserve imported subscription routing (#462)
- feat(agent-context): add native context compaction (#463)
- feat(theme): add dark mode with toggle in Settings > General (#399)
- feat(workspace): add whole-window find bar (#466)
- fix(electron): strip trailing newline from electron path.txt (#464)
- style(workspace): refine notebook composer appearance (#465)
- fix(workspace): harden whole-window find overlay (#467)
- fix(compute): source Bash config before remote workloads (#470)
- chore(release): bump version to 0.7.3
Full Changelog: v0.7.2...v0.7.3