Repository navigation
AIPOCH Open-Science v1.3.1
Built-in web search and page reading, an interactive citation graph, MCP access without opening the app, Lab models and organization cases, fresher provider catalogs — plus a fix for updating from 1.3.0.
AIPOCH Open-Science is an open-source, local-first, model-agnostic 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.
v1.3.1 follows v1.3.0 with a connectivity-and-reliability release. The built-in Open-Science Agent can search the web and read public pages, citing sources that stay with the conversation. Literature gains an interactive citation graph that re-centers on any related paper. External agents can now connect through MCP without opening the app first, with setup guides for more than 30 agents and clients. Lab account sign-in unlocks organization models and shared research cases, and provider model lists can be refreshed from Settings with failures that explain themselves. Notebook runs check memory before they start, Windows installation and update paths are more resilient, Linux AppImage updates recognize the app's own processes, and a 1.3.0 updater issue that could block the update is documented with a one-time manual step.
✨ Highlights
- Web search and page reading for the built-in agent. The Open-Science Agent can search the web and read public pages. Search approvals show the query and its filters in a readable card, and the sources it cites stay with the conversation across restarts, branches, and research packages. (#4448, #4471)
- Interactive citation graph. A paper's references and citing papers appear as a graph you can explore. Re-center the graph on any related paper, filter by direction, and return to the original paper. (#4418, #4467)
- Use your agent through MCP without opening the app. When an external agent connects to Open-Science through MCP, the background service starts on its own. The setup guide now covers more than 30 agents and clients. (#4470, #4464)
- Lab models and organization cases. Sign in with a Lab account to use the models your organization assigns, and search, import, or share research cases with your teammates. (#4447)
- Fresh model lists and clearer provider errors. Refresh a provider's model list from Settings. When a provider request fails, the conversation names the cause and offers the matching next step. (#4415)
⚠️ Scope & Safety
- Updating from 1.3.0 may need a manual installation. The 1.3.0 updater can mistake one of the app's own background processes for another program and refuse to install. This release cannot change the updater that is already running in 1.3.0. If the update reports that another process is using the installation, fully quit Open-Science and its background service, then install the downloaded package manually. Your profiles and research data stay as they are. Updates that start from 1.3.1 are not affected. See the known-limitations register. (#4454)
- Web search and page reading contact outside services. Search queries go to Exa, a third-party search service, without an account. Reading a page contacts that page's website. Both tools follow your permission mode, and stopping a request cannot retract what was already sent. See data recipients. (#4448)
- Lab sign-in changes where requests go. With a Lab account, requests to Lab models go through the Lab server you sign in to. Sharing a case uploads that session's research package to that server for your organization. Review the package before you share it, and review an organization case before you import it. (#4447)
- Connecting an agent keeps the service running. Connecting an external agent starts the background service, and the service keeps running after the agent disconnects. Stop it with
open-science stopwhen you no longer need it. (#4470)
🚀 New Features
- The built-in Open-Science Agent can search the web and read public pages. It cites its sources, and those sources stay available offline, across branches, and in research packages that allow it. (#4448)
- Web search approvals show the query, time range, result limit, and domain filters in a readable card. (#4471)
- Explore a paper's citation network as an interactive graph. Choose All, References, or Cited by, follow arrows that show citation direction, re-center on any related paper, and return to the original paper. (#4418)
- External agents connected through MCP start or reuse the Open-Science service for your profile when they connect, so you no longer need to open the app first. (#4470)
- The README and CLI guide include an external-agent quick start, with setup options for more than 30 agents and local clients and a first call that confirms the connection. (#4464)
- Sign in with a Lab account to use your organization's models and track their usage. Search your organization's research cases, import verified cases, and upload a session to share it with a share code. (#4447)
- Refresh a provider's model list from Settings. A failed refresh keeps the previous list. (#4415)
- Image services can run an optional test generation before you save them. Saving a new image service selects its first model when no default model is set. (#4455)
🔧 Improvements
- Provider failures name their cause in the conversation, including authentication, quota, rate limits, unavailable models, connection loss, interrupted responses, and context limits, and offer the matching recovery action. (#4415)
- Notebook runs check available memory before they start. When the system is low on memory, a run is not started and you can retry it yourself. From Compute, you can choose a session and release its Notebook resources while active work stays protected. (#4374)
- The project home appears sooner at startup. Session details load when you open a session instead of all at once. (#4434)
- When the saved research folder is missing, the startup recovery dialog shows that folder's path and asks you to reconnect the drive. (#4472)
- First-run onboarding on Windows no longer stops at the optional WSL step. You can set up WSL2 Bash later in Settings → Runtimes. (#4463)
- Windows installation failures show their diagnostics faster. (#4457)
- Citation graph arrows meet their target papers, and the off-screen hint takes less space on small screens. (#4467)
- Lab sign-in closes the authorization tab when it finishes, and the Lab account's connection settings are grouped together. (#4475)
🐛 Bug Fixes
- In-app updates no longer treat the app's own background log process as another program that blocks installation. (#4454)
- Windows uninstall and update no longer fail when the settings file is large. (#4469)
- After a reconnect, sessions keep unsent drafts, load long branched histories, and accept answers to questions restored after a crash. Renaming or pinning a session no longer overwrites newer changes. (#4460)
- A session whose details failed to load no longer retries that load over and over. (#4452)
- New agent sessions no longer show repeated literature warnings while they start. (#4462)
- Deleting an imported session on Windows removes its files and records completely. (#4459)
- A Notebook session opened on its own loads the agent's conversation history instead of starting without it. (#4461)
- Research films no longer wait on quality reviews that do not affect playback. (#4468)
- AppImage updates on Linux no longer mistake the app's own keepalive pipe for a process that blocks installation. (#4479)
📦 Install
Desktop client. Requirements: macOS 12+ (Apple Silicon or Intel), Linux x64 or ARM64, 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 updates itself in place.
Download the appropriate package from the Assets section below, or from the official download page. On macOS, the desktop app is also installable via Homebrew: brew install --cask open-science.
| Platform | Package |
|---|---|
| macOS (Apple Silicon) | DMG for ARM64 |
| macOS (Intel) | DMG for x64 |
| Linux | AppImage or Debian package for x64 / ARM64 |
| Windows | Code-signed installer for x64 |
Updating from 1.3.0. If the 1.3.0 updater reports that another process is using the installation, fully quit Open-Science and its background service, then install the downloaded 1.3.1 package manually. Updates that start from 1.3.1 are not affected. Profiles and research data are preserved.
Command line via Homebrew (macOS and Linux). The standalone CLI and local Web UI live in the official tap — bundled Node runtime, no separate Node.js needed, and upgrades through the normal brew flow:
brew install aipoch/tap/open-science
open-science start --no-open
open-science url
Command line via npm. With Node.js 22.13+, install the CLI (and Task SDK) globally; updates then follow your normal npm upgrade flow:
npm install --global @aipoch/open-science
open-science --help
Command line standalone (all platforms, no toolchain). The Assets below also include standalone CLI archives — macOS and Linux arm64/x64 tarballs and a Windows x64 zip — with a bundled Node runtime, so they install and run without Homebrew, npm, or a local Node.js. Linux packages additionally arrive as standalone .deb and .rpm installers.
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 notarized differently and may require approval through macOS Privacy & Security.
Windows — first launch. Official release builds are code-signed; Windows recognizes the publisher and no SmartScreen warning appears. 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: everything shipping in v1.3.0, plus built-in web search and page reading with source citations, the interactive citation graph, MCP access that starts the service on its own with a 30+ agent setup guide, Lab organization models and case sharing, live provider model-catalog refresh with structured failure causes, Notebook memory-pressure admission and resource release, and the fixes described above.
- 🚧 Partial: remote compute covers direct SSH, Slurm, and user-connected Kubernetes, Hugging Face Jobs, Modal, and Ray targets, each requiring user-provisioned accounts, software, and permissions; provider choice remains constrained by the active framework's endpoint compatibility; WSL2 Bash remains an opt-in Windows x64 preview; the skills commons ships marketplace discovery but not cross-machine forking or user-facing version pinning; reproducibility ships replayable per-version verification, session replay, and RO-Crate packaging, while solver-exact equivalence remains open; review is opt-in and record-scoped.
- 🗺️ Roadmap: a unified model gateway, cross-machine skill forking and version pinning, stronger sandboxing and credential isolation, and collaborative research workflows.
🐢 Known Limitations
- Updating from 1.3.0 may require one manual installation: the 1.3.0 updater can mistake the app's own background log process for another program using the installation. Fully quit Open-Science and its background service, then install the downloaded 1.3.1 package manually. Updates that start from 1.3.1 are not affected. See the known-limitations register. (#4454)
- Replay reproduces recorded steps rather than re-executing them, and code reconstruction is LLM-generated — solver-exact equivalence remains open, and replayable verification is the deterministic check (starting one from the web client currently requires the desktop app).
- PDF annotations are bound to the file version they annotate and do not sync across machines.
- AI-assisted outputs — study snapshots and the reviewer — reflect the configured main model and the record they cover; they are aids, not validation of citations, units, statistics, or methods.
- Network sandboxing covers the notebook and compute runtimes, not the whole system — on Windows it applies only after the sandbox's one-time administrator setup, and Strict mode narrows Notebook egress to a manually managed whitelist whose built-in destinations deserve a review before you rely on them.
The full limitations and security register lives in docs/known-limitations.md.
🙏 Acknowledgements
Thanks to @wen2zhou, @nasus2002, @ShuangChenyue, @liguohhh, @guch96, @liyeshunag and everyone in Discord, X, and Discussions.
Full Changelog: https://github.com/aipoch/open-science/commits/v1.3.1