Repository navigation
AIPOCH Open-Science v1.3.0
Scheduled research tasks, image generation and editing, editable presentations, research-film replay with independent branch outputs, and smarter agent questions — plus a faster, more reliable core.
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.
v1.3.0 follows v1.2.2 with a feature release built around automation and creation. Scheduled tasks launch sessions and Notebook runs on a recurring or one-time cadence, each with its own record in the workspace. Images can be generated with the configured provider and iteratively edited, with every variant kept as a new version. PowerPoint files open for in-place text editing with formatting while the slide structure stays intact. Research films gain operation replay with independent branch outputs, open from the session menu, and explain token usage up front. When the agent asks for input, options now lay out their trade-offs. CodeBuddy subscribers can sign in with a regional China or International account and pick from the models it can access. Underneath, long session histories persist incrementally, stalled films and package transfers recover on their own, transient film failures retry automatically, and Windows installs and updates get leaner and faster.
✨ Highlights
- Scheduled research tasks. Set up recurring or one-time schedules that launch sessions and Notebook tasks on their own — each run opens in the workspace with its own record, and you stay in control with clear run history. (#4380)
- Image generation and editing. Generate images from a prompt with your configured provider, then iterate: each edit keeps a new version, so you can compare, revisit, and reuse earlier variants across the project. (#4424)
- Editable presentations. PowerPoint files open for editing: revise text with formatting — fonts, sizes, colors, alignment — and keep the file's structure intact, all inside the app. (#4392)
- Research films, replayed. Film operations can be replayed onto new branches with independent outputs, films open straight from the session menu, and the film dialog explains token usage before you generate. (#4389, #4414)
- Smarter agent questions. When the agent needs your input, it now lays out the trade-offs between the options it offers, so you can decide with the consequences in view. (#4357)
⚠️ Scope & Safety
- Scheduled tasks run with your project's permissions. A schedule launches its session or Notebook run without a further confirmation while the app is running. The run acts with the folders, credentials, and network access already granted to that project, and it consumes your configured provider quota. Review the schedule's target, cadence, and the project's grants before enabling it, and pause or delete schedules you no longer want. See the scheduled-task boundary in the known-limitations register. (#4380)
- Image generation and editing disclose prompts and source images to your configured provider. Generation requests and the images used as their source are transmitted through the provider connection you configured, and cancelling or deleting locally cannot retract bytes the provider has already received. Review the provider's data terms before generating from sensitive material. See data recipients. (#4424)
🚀 New Features
- Create scheduled tasks that launch a session or run a Notebook on a recurring or one-time cadence: each scheduled run opens in the workspace as its own record, with run history kept per schedule. (#4380)
- Generate images from prompts through your configured provider and edit them with follow-up instructions: every generation and edit stores a new artifact version, and earlier variants stay available for comparison and reuse. (#4424)
- Edit PowerPoint presentations in place: select and revise text with fonts, sizes, colors, and alignment while the slide structure is preserved. (#4392)
- Replay research-film operations onto a different branch and get an independent output without touching the original film. (#4389)
- Open generated research films directly from the session menu, and see an up-front explanation of how many tokens film generation will use. (#4414)
- Monitor running Notebook processes from the app and stop the ones the app owns when they are no longer needed. (#4399)
- Document and figure skills come ready to use: their supporting libraries are preinstalled into the managed environment on first use instead of failing mid-task. (#4411)
- Sign in to a regional CodeBuddy subscription (China or International) from Settings: the managed CLI handles authentication and discovers the models your account can access, and model selection is confirmed by the CLI. (#4442)
🔧 Improvements
- Long conversation histories persist incrementally, so large sessions save continuously, reload faster, and stay responsive while they grow. (#4378)
- The app reaches the project home sooner at startup by trimming work that blocked the first screen. (#4398)
- Stalled film generation, research-package transfers, and background deliveries recover on their own instead of waiting for a restart. (#4372)
- Windows installs and updates use a leaner, more reliable packaging pipeline, and update readiness checks are bounded so they cannot hang the flow. (#4420, #4417)
- Recovery scheduling for compute targets runs on demand instead of on a fixed timer, reducing idle overhead. (#4375)
- Skill references align with the public OpenCode 2.x skill names, so imported skills resolve as expected. (#4355)
- External PDFs recover their native structure more reliably during analysis and translation, the reader consolidates its toolbar, and image-support limits explain themselves instead of failing silently. (#4361, #4360, #4435, #4407, #4394, #4400)
- The composer keeps send controls reachable in narrow layouts, shows context usage accurately, and reveals the scroll-to-latest arrow when you hover the conversation. (#4367, #4403)
- First paint of long conversations is capped so the window opens quickly even with many loaded messages. (#4376)
- Windows updates verify the installer signature resiliently under system load, and the installer notices a running app sooner instead of relying on fixed waits. (#4445)
- Film generation retries a transient model failure automatically once after a short delay, sharing a small retry budget across the generation phases. (#4446)
- Windows updates close processes that hold installation files natively instead of spawning a helper per process, so the update handoff finishes sooner and more reliably. (#4450)
🐛 Bug Fixes
- Sessions keep their correct numbering when records are repaired, and journal replay no longer repeats work or rescans history unnecessarily. (#4368, #4432, #4428)
- Agent runtimes release their resources across agent lifecycles, so repeated runs do not accumulate overhead. (#4396)
- Skill reloads no longer interrupt in-progress session creation. (#4427)
- Agent handoffs preserve their metadata even when tool results are truncated. (#4413)
- Artifact downloads through the local CLI no longer time out while data is still arriving. (#4430)
- Remote task requests and downloads honor the caller's lifetime, so abandoned operations stop promptly. (#4390)
- Terminal Ray jobs are removed after harvest instead of lingering on the target. (#4386)
- ACP resource links are accepted again. (#4381)
- Science packages open the available imported film mode instead of failing. (#4419)
- Previews keep their filename identity in narrow surfaces, run marks land at the right positions, and the project switcher keeps its scrolling inside the list. (#4401, #4350, #4405)
- Notebook frame transcripts paginate steadily, and Markdown footnotes keep their scroll contained. (#4370, #4391)
- Film generation distinguishes its model deadline from a user cancellation, names each timeout cause clearly, and labels checkpoint retries accurately. (#4444)
📦 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 |
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 v1.2.2, plus scheduled session and Notebook tasks, image generation with versioned editing, in-place PowerPoint text editing, research-film replay with independent branch outputs, trade-off explanations in agent questions, regional CodeBuddy subscription sign-in, Notebook process monitoring and stop, preinstalled document and figure skill libraries, 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
- 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 @cyj1116, @wen2zhou, @fengdwx, @roxi3906, @BobbyLin7, @daanveer-tech and everyone in Discord, X, and Discussions.
Full Changelog: https://github.com/aipoch/open-science/commits/v1.3.0