aipoch/open-science 发布 v0.26.0：新增 HPC Slurm 执行模式与文献引用库

Open Science by AIPOCH is an open-source, local-first, model-agnostic AI research workbench for macOS, Windows, and Linux, with scientific agents, Python/R notebooks, data connectors, and reproducible provenance. Release v0.26.0 published_at 2026-09-07T01:13:01Z # Open Science v0.26.0 > A release that pairs HPC-class compute with a literature workspace: remote compute hosts gain a per-host Slurm execution mode alongside direct SSH, and a new reference library organizes references, PDFs, and citations — alongside new providers and fresher model catalogs, readable tool activity, quieter default permissions, and smoother streaming. 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. v0.26.0 connects the workbench to HPC clusters and gives literature a home. Remote compute hosts now select a per-host execution mode — direct SSH or Slurm — so notebook runs can submit as Slurm jobs on HPC-partitioned clusters, with durable submission, polling, recovery, cancellation, and cleanup, and a guided Compute Environment Setup skill that hands the cluster's exact setup, repair, and removal steps to the user or administrator. A new reference library organizes references, downloaded PDFs, and citations into collections with identifier-aware imports, duplicate comparison and bulk merge that preserves attachments, full-text PDF attachment through public open-access sources, and citation formatting with provenance. Apodex joins the built-in providers and the latest OpenAI and Anthropic models — GPT-6 Astra and Claude Fable 5.1 — are selectable out of the box. Artifact writes and notebook controls render as readable summary cards in messages and approvals, safe default permissions expand to cover routine read-only inspections so everyday work runs with fewer interruptions, and streaming output, long-session resource bounds, and update handling are smoother throughout. ## ✨ Highlights - **Run on HPC clusters with Slurm.** Remote compute hosts gain a per-host execution mode — direct SSH or Slurm — so notebook runs can submit as Slurm jobs, with durable submission, polling, recovery, cancellation, and cleanup, and a guided setup skill that hands your cluster's exact setup steps to you or your administrator. (#2238) - **A reference library for your literature.** Import references by identifier or file, organize them into collections, link them to projects, compare and bulk-merge duplicates, attach full-text PDFs from public open-access sources, and format citations with provenance. (#2236) - **New providers and fresher model catalogs.** Apodex joins the built-in providers, and the latest OpenAI and Anthropic models — GPT-6 Astra and Claude Fable 5.1 — are selectable out of the box. (#2147, #2175) - **Tool activity you can read.** Artifact writes and notebook controls (runtimes, restarts, bindings, kernel state) render as compact summary cards in messages and approvals instead of raw JSON. (#2224) ## 🚀 New Features - **Slurm execution mode** — per-host direct-SSH or Slurm execution for remote compute hosts across the desktop app, web access, and the Host SDK, with an execution-mode selector in remote-compute settings, scheduler-owned job naming for safe recovery, and a user-managed Compute Environment Setup skill with exact setup, repair, and removal instructions. (#2238) - **Literature reference library** — collections, project links, inbox acceptance for downloaded PDFs, trash, identifier-aware imports, citation formatting with artifact provenance, side-by-side duplicate comparison, and bulk merge that preserves attachments; find and attach full-text PDFs through Europe PMC, PMC, OpenAlex, arXiv, and Unpaywall — all applicable sources are looked up in parallel — with optional credentials. (#2236, #2234, #2262) - **Apodex provider** — official Apodex models (`apodex-1.1` and `apodex-1.1-mini`) with 262,144-token contexts, routed correctly for each agent framework. (#2147) - **Latest OpenAI and Anthropic models** — GPT-6 Astra (1,050,000-token context) and Claude Fable 5.1 (1,000,000-token context) with reasoning-effort support. (#2175) - **Tool summary cards** — compact, readable cards for artifact writes and common notebook controls in messages and permission approvals. (#2224) - **Expanded safe default permissions** — routine read-only inspections (notebook runtimes and state, memory queries, package inventories) and approved-plan progress updates no longer interrupt with approval prompts; grants remain visible and revocable in Settings. (#2225) - **Preview content context menus** — right-click inside previews: copy path, download, or save local files as artifacts; inspect provenance or return to the originating context for managed artifacts. (#1862) - **Expandable skill document in approvals** — skill-load approvals expand to render the full skill document before you approve. (#2056) - **Task API session configuration** — inspect and update a session's model, reasoning effort, memory, and compute-host selection from the CLI, HTTP API, or SDK without submitting a prompt. (#2121) - **Asset-heavy skill imports** — large skill bundles with tens of thousands of files import reliably and quickly. (#2138) ## 🔧 Improvements - Streaming responses render more smoothly: per-frame overhead, double pacing, and bottom-follow flicker are gone. (#2141) - Long sessions stay fast: session resource usage is bounded, and immutable file verification is cached for quicker version history. (#2167, #2158) - Approved session plans survive restarts and new attempts — the active plan is rebuilt from durable session state instead of failing with a continuation error. (#2152) - Update handling is more forgiving: cancellation is honored, offered updates are preserved, and platforms without an installer get an explanation with a manual-download link. (#2245) - Long sessions stay responsive: prompt dispatch no longer rescans the full transcript, and handoff, preview, and window reads are bounded. (#2256, #2259) - Notebook environments can use automatic package mirrors. (#2127) ## 🐛 Bug Fixes - **Notebook and runtimes** — protected Windows REPL startup is restored (#2122); Windows R detection goes through Rscript (#2212) and standard launches preserve stdin (#2118); managed runtime state is isolated (#2217); environment repair and package outcomes are reported correctly (#2214); reused cell languages are honored and run inputs preserved (#2206, #2207); abandoned code write streams are recovered (#2205); installer caches stay inside managed runtime storage (#2190); and kernel activity settles after run write failures. - **Files, artifacts, and previews** — previews stay synchronized with publication (#2191, #2198); text reads retry during publication (#2154); managed resources stay pinned across pagination (#2197); finalization recovers in multi-message turns (#2187); standalone attachments return correctly after publication (#2196); CSV counts and Unicode title search are corrected (#2178); and completion rows no longer flash on height changes. (#2153) - **Sessions and persistence** — branch context survives failed forks (#2213); newer metadata wins during lazy hydration (#2209); streamed chunks survive clock rollback (#2159); deletion boundaries and compensation failures are preserved (#2193, #2165); retained artifacts stay readable (#2160); stalled provider deletions time out (#2149); concurrent project deletions are isolated (#2172); and message snapshots are durable. (#2161) - **Agents, providers, and delegation** — provider state survives async completions (#2232) and validation scopes to model targets (#2136); tool content and response semantics are preserved (#2233); agent-process ownership holds across recovery and permission waits (#2219); delegation results and lifecycles are preserved (#2220); specialist recovery and approved capabilities are kept (#2226, #2221); and custom provider URLs are validated and redirects rejected. (#2124, #2126) - **Skills, connectors, and memory** — skill packages keep their integrity across edits and imports (#2227, #2180); connector data is preserved and credential updates recover (#2229); connector response bodies are released and call budgets enforced (#2182); and memory edits are guarded while short search terms keep working. (#2185) - **Updates, platform, and workspace** — Windows installers preflight locked update targets (#2155); the Linux application menu stays visible (#2139); startup waits are bounded and quits are guarded until the runtime hands over (#2166, #2202); live proxy state is restored after a failed apply (#2169); diagnostics logging is hardened and statuses refresh (#2249); hover previews give clearer interaction feedback (#2263); and accessibility, contrast, and keyboard navigation are restored. (#2176, #2177) - **Sessions and archiving** — archiving waits until a session's compute jobs are truly finished, and archive and restore preconditions use durable versions so delayed commands cannot slip through; rejected workspace deletions surface their error inside the dialog with an explicit retry. (#2258) - **Storage, credentials, and remote access** — data-location migration keeps references and recovery semantics intact (#2257); credential recovery is restored and saved state reconciles (#2260); remote access recovers from persistence and probe failures (#2254); and usage records keep their execution identity and recover. (#2248) ## ⚠️ Breaking Changes - The public Task SDK contract removes legacy session-plan continuation fields. Approved plans now rebuild from durable session state for every attempt; clients that still read the removed fields must use the current session and plan APIs. (#2152) ## 📦 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, composer session references (`#`) with turn-scoped read access, reversible archiving with keyboard undo, project pinning, a project quick switcher listing other active projects with title and description previews and fuzzy search once the list grows, collapsed long user messages, persistent side conversations with advisories injected into running main turns, generated and editable session details, session hover previews in the sidebar, and SQLite-indexed summary-first session startup; in-app sandboxed previews for source links in agent responses; selectable Claude Code, OpenCode, Codex, and CodeBuddy agent frameworks (CodeBuddy app-managed and login-free) behind a shared provider turn-adapter interface, with agent shell commands routed through the app's own execution path on every framework; text, image, and PDF annotations that send selected context into conversations with click-to-reveal evidence in the source document, a session reading context that links up to three PDFs the agent can read, page through, and search, with agent configuration change markers in the timeline; opt-in persistent agent memory with project-scoped categories recalled across sessions and managed from Settings; production subagent delegation with durable messaging, restart recovery, structured output, artifact and review evidence, camelCase delegation APIs, and a per-session delegation switch in the composer's agent controls; review-gated session plans with CLI plan controls; compact summary cards for artifact writes and notebook controls in messages and approvals; a unified composer lane with a session-scoped message queue, unified draft undo and redo history, active-specialist emphasis, and mid-turn Send now through native follow-up steering; hot-switching of compatible models and providers; multi-provider model configuration including Apodex, NVIDIA Build with a curated agent-capable catalog, the latest OpenAI and Anthropic model catalogs, Tencent Coding Plan and Token Plan subscription providers, an xAI OAuth subscription, a dedicated Vision model selector with persistent evidence relay, custom token limits, and a consolidated Scenario models card; per-model-call usage details with a per-call context-window chart; a token usage dashboard with persisted per-run attribution that includes auxiliary model calls outside the main conversation; centralized credential management for GitHub tokens, connector keys, and connector sign-ins with guided recovery, plus device-wide shared credentials (API keys, access tokens, and OAuth sign-ins) that custom connectors bind to as environment variables, headers, or sign-ins; a configurable reviewer model policy with an isolated review runtime, durable assessment snapshots, and empty initial checks, with bounded evidence reads and correction round-trips that preserve your responses and the reviewer model identity; context-window composition insights with compaction boundaries; persistent Python/R/REPL kernels with bounded run-history payloads, a read-only live variable browser for running kernels, live variable-name suggestions in the shared terminal, a docked Variables pane in wide notebook previews, notebook and compute network access limited to Open Science defaults and user-approved domains with in-conversation approvals for blocked destinations (enforced out of the box on macOS and Linux; on Windows after the sandbox's administrator setup), a status-aware Notebook network protection panel in Settings → Runtimes, safe managed-runtime reinstall from Settings, and a global toggle for agent-created runtime environments, immutable generations for files created or modified by notebook and compute runs, separated approval and execution state, cross-run dependency tracking, package-installation progress, progressive history loading, terminated read-only previews, session-scoped remote compute execution with key or password authentication (including Windows), a per-host execution mode (direct SSH or Slurm) with durable submission, polling, recovery, cancellation, and cleanup plus a guided Compute Environment Setup skill, harvest ceasing, and remote file-staging cleanup, with remote compute jobs that survive restarts and crashes through durable operation receipts and automatic recovery; a user terminal shared with the agent; app-managed and bring-your-own environments for Python and R with notebook workload caches under the configured data-storage location, and data-location migration that shows the target drive's capacity and estimated copy size; immutable, session-scoped artifact versions with checksummed content, producer code, execution history, exact input references, environment inventory, producing message-branch context, version-scoped reviewer evidence, and on-demand LLM-driven code reconstruction, with allowlisted text artifacts and uploads editable as raw text where every save publishes a new comparable version; a literature reference library with collections, project links, identifier-aware imports, duplicate comparison and bulk merge, open-access full-text PDF attachment through Europe PMC, PMC, OpenAlex, arXiv, and Unpaywall with parallel multi-source lookup, and citation formatting with artifact provenance; rich in-app previews for scientific data, documents, images, source code, molecular structures, and notebook history with right-click tab and content actions and full-screen mode; file attachments up to 10 GB with streaming upload; skills with conversational creation, import (including large, asset-heavy bundles), marketplace browsing, and explicit `/` selection, plus provenance-aware figure workflows for styling, composition, and paper narratives; 24 built-in research connectors plus custom MCP servers with import/export of standard MCP client configurations; durable scoped permissions with allow grants, expanded safe seeded defaults that cover routine read-only inspections and approved-plan progress updates, and a restore-defaults action for the baseline; remote-access pairing; interface localization in German, Spanish, French, Chinese (Simplified and Traditional), Japanese, Korean, and Russian; and auto-update with prominent update reminders and localized release notes. - 🚧 **Partial:** R remains managed-only; provider choice remains constrained by the active framework's endpoint compatibility; remote compute covers direct SSH and Slurm (cloud-GPU submission is not built yet); skills remain local (no hosted public discovery commons); reproducibility stops at preserved evidence — captured file generations are groundwork, not yet automated reruns; the literature library is a first-generation workspace; and review is opt-in and record-scoped. - 🗺️ **Roadmap:** a unified model gateway, a hosted public skills and specialist discovery commons, cloud-GPU execution, stronger sandboxing and credential isolation, and collaborative research workflows. ## 🐢 Known Limitations - **Remote compute covers direct SSH and Slurm.** Cloud-GPU submission is not built yet. - **The literature library is a first release.** Collections, imports, duplicate merging, open-access PDF attachment, and citation formatting ship today; deeper reference-aware research workflows continue to evolve. - **Editable artifacts cover text formats.** Markdown, plain text, scripts, and source code are editable as raw text; binary formats such as documents, images, and notebooks stay read-only, and editing always publishes new versions rather than rewriting history. - **Network sandboxing covers the app's notebook and compute runtimes, not the whole system.** Processes you launch outside these runtimes are not subject to the domain allowlist. On Windows, the boundary applies only after the sandbox's one-time administrator setup; until then, notebook and compute code runs without it. - **Immutable file generations preserve earlier results, but deterministic reruns are still ahead.** Captured generations are the groundwork; portable environment locks and full-fidelity session replay remain open. - **R is managed-only.** A bring-your-own R interpreter path is 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. - **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, cancellation, and session configuration 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.** Directory-level file access control is not built yet; notebook network sandboxing is the first enforced network boundary. - **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, @LiuZhaoyan, @wen2zhou, @roxi3906, @daanveer-tech, and everyone in [Discord](https://discord.gg/zxQAYjReRv), [X](https://x.com/aipoch_ai), and [Discussions](https://github.com/aipoch/open-science/discussions). --- **Full Changelog**: https://github.com/aipoch/open-science/commits/v0.26.0 README (https://github.com/aipoch/open-science/blob/main/README.md): AIPOCH Open Science Open-source, local-first, model-agnostic AI research workbench for reproducible science. Open Science is an open-source, local-first, model-agnostic AI research workbench developed by [AIPOCH](https://aipoch.com/open-science) for scientists and researchers. It enables reproducible, inspectable research with scientific AI agents, Python and R execution, scientific data connectors, and cross-platform support for macOS, Windows, and Linux. Create a project, describe your research goal in plain language, and let the agents read files, search the web, run code, query scientific data sources, and produce reports, tables, and figures with traceable provenance—all in one workspace. Open Science supports computational and data-intensive research across disciplines, including machine learning, statistics, life sciences, chemistry, materials science, physics and environmental science. It supports the research process from literature review and hypothesis development to code execution, data analysis, simulation, visualization, and the production of traceable research outputs. > 💡 **[Open Science v0.26.0 released](https://github.com/aipoch/open-science/releases/latest)** _(last updated September 2026)_. Open Science v0.26.0 brings HPC-class compute and a literature workspace: remote compute hosts gain a per-host Slurm execution mode alongside direct SSH, and a new reference library organizes references, PDFs, and citations with identifier-aware imports, duplicate merging, open-access full-text attachment, and citation formatting. Apodex joins the built-in providers alongside the latest OpenAI and Anthropic models, notebook tool calls become readable summary cards, and smoother streaming, quieter default permissions, and a broad set of fixes land throughout. See the [latest release notes](https://github.com/aipoch/open-science/releases/latest) for full details. ## Table of Contents - [Quick Start](#-quick-start) - [Product Tour](#product-tour) - [Benchmark Performance](#benchmark-performance) - [Why Open Science](#why-open-science) - [Design Principles](#design-principles) - [Core Capabilities](#core-capabilities) - [Model Providers](#model-providers) - [Data, Permissions, and Trust](#data-permissions-and-trust) - [Project Status](#project-status) - [Development & Packaging](#development--packaging) - [Roadmap](#roadmap) - [Relationship to the AIPOCH Ecosystem](#relationship-to-the-aipoch-ecosystem) - [What This Is Not](#what-this-is-not) - [Frequently Asked Questions](#frequently-asked-questions) - [Get Involved](#get-involved) - [License](#license) - [Star History](#star-history) ## 🚀 Quick Start Get Open Science running in three steps: download the installer for your platform, complete the guided first-run setup, and create a research project. ### 1. Download the app Open the [latest release](https://github.com/aipoch/open-science/releases/latest), expand **Assets**, and choose the installer for your computer: | Your computer | Choose | | ----------------------------------- | ---------------------------------------- | | macOS — Apple Silicon (M1 or newer) | The macOS DMG for Apple Silicon / ARM64 | | macOS — Intel | The macOS DMG for Intel / x64 | | Windows x64 | The Windows x64 installer | | Linux x64 | The Linux x64 AppImage or Debian package | Review the assets and verification information published on the release page. See [Verifying your download](SECURITY.md#verifying-your-download) before installation if you need to validate a package. > If macOS or Windows shows an unidentified-developer or unknown-publisher warning, verify that the package came from the official Releases page before continuing. ### 2. Complete first-time setup The first launch has five guided steps: 1. **Environment** checks compatibility, app storage, secure credential storage, and network access. 2. **Data location** chooses where large artifacts, notebooks, uploads, and environments are stored. 3. **Agent runtime** selects and prepares Claude Code, OpenCode, Codex, or CodeBuddy. App-managed runtimes can be installed without requiring Node.js, npm, or an administrator password. 4. **Model provider** connects and tests the model you want to use. Choose a built-in provider, a custom gateway, or an existing Claude or Codex subscription login. 5. **Notebook runtime** optionally prepares app-managed Python and R environments or enables detected and manually registered interpreters for either language. Host compatibility, storage, and network checks Provider, API Key, endpoint, and model validation Notebook execution is optional. Every required environment and agent-runtime check must pass before `Continue` becomes available, and the model connection must pass before setup finishes. Notebook and data-location settings can keep their defaults and be changed later in Settings. While a kernel is running, a Variables view can inspect the live Python or R namespace — names, types, shapes, and previews — read-only, refreshed after each execution. ### 3. Start a research project 1. Click **New project** and give the project a stable research name and optional description. 2. Open a session and describe the goal, input data, constraints, desired outputs, and how the result should be checked. 3. Attach source files, select a verified model, and choose an approval mode. 4. Send the task. Inspect the agent's tool activity, approve sensitive actions, and open generated artifacts in the preview panel. 5. To explore a different direction, edit an earlier user message and resend it on a new branch; use the message revision controls to return to either path. 6. Open an artifact's **Provenance** view to inspect its versions and the available evidence behind the selected result. 7. Continue the work in later sessions. Use `@` to reference an existing project file and `/` to explicitly select an enabled skill. > Screenshots in this README illustrate the workflow. Labels, catalogs, and other interface details may differ from the version you install. ## Product Tour Open Science organizes research into projects and sessions so that every result can stay connected to the evidence that produced it. The sections below walk through the workspace, artifact provenance, previews, scientific skills, and data connectors. ### One workspace from task to traceable artifacts Projects keep related sessions, uploads, generated files, and preview state together. The conversation records the agent's answer and the commands, file reads, edits, searches, and connector calls that produced it. Each generated artifact is stored as an immutable, checksummed version. Its **Provenance** view exposes the evidence Open Science could verify at creation time: producer code and execution history, referenced inputs, an observed environment inventory, the producing conversation branch, and any version-scoped reviewer findings. Missing evidence is shown as unavailable instead of being guessed. Uploads and generated files organized by project and session Native previews keep data and the research history side by side Generated reports, figures, and tables remain attached to the session and are also collected in the project file library. Preview tabs keep the active result visible as the panel changes size, and long names preserve their identifying suffix and extension. Open Science previews common scientific data, PDFs, Office documents (DOCX, XLSX, PPTX), images (with zoom and pan), source code with syntax highlighting, molecular structures and reactions, and Notebook history. Preview limits do not truncate the underlying file—the full artifact stays available to the agent and external tools. Use `Cmd/Ctrl+F` to search transcripts, Notebook output, and rendered pages across the workspace, or `Cmd/Ctrl+K` to open the project-scoped command palette. A dark mode rounds out the workspace: toggle the theme in **Settings → General** and the whole shell, transcript, and renderer palette switch without a flash. The interface is also available in German, Spanish, Chinese (Simplified and Traditional), Japanese, Korean, French, and Russian with a runtime language switcher in Settings. ### Branch a conversation without losing the original Edit a completed user message to resend a revised prompt from that point. Open Science creates a new message branch instead of deleting the turns that followed, and revision controls let you move between the original and alternative paths. Branch selection, tool activity, attachments, and generated artifacts persist across project switches and restarts. Provenance remains tied to the exact branch that produced each artifact version, so exploring a different hypothesis does not blur the record of the earlier result. ### Scientific skills and data connectors Open Science includes a growing catalog of **22 featured**, file-based research skills: AlphaFold2, Boltz, Borzoi, Chai-1, Customize, DiffDock, Environment & Packages, ESM-2, ESMFold2, Evo 2, Figure Composer, Figure Style, Indication Dossier, LigandMPNN, Literature Review, OpenFold3, Paper Narrative, ProteinMPNN, scGPT, scvi-tools, SolubleMPNN, and **Remote Compute (SSH)** for submitting and harvesting long-running jobs on remote HPC clusters. You can create personal skills, upload `SKILL.md`/ZIP/`.skill` packages, preview and import compatible skills from GitHub with optional authenticated access, or import skills already installed in your global agent directories. The agent can also request a package import from a session attachment or a public GitHub URL, with an app-owned preview and confirmation step before anything is written. Enabled skills can be selected directly in the composer with `/`. It also includes **24 built-in** research connectors: Literature Graph, PubMed, bioRxiv, Genes & Ontologies, Genomes, BioMart, Variants, Human Genetics, Clinical Genomics, Structures & Interactions, Protein Annotation, Expression, Omics Archives, CellGuide, Regulation, RNA, Chemistry, ChEMBL, ZINC, Molecule Viewer, Clinical Trials, Drug Regulatory, Cancer Models, and Research Resources. Built-in and custom connectors remain behind the permission system, with per-tool `Always allow`, `Ask each time`, and `Block` controls. The installed app shows the current skill, connector, and tool catalogs. Readable, reusable research skills Scientific databases exposed as permissioned agent tools ## Benchmark Performance ### 🏆 #1 on BiomniBench-DA Public 50 Open Science achieved the highest ranking score in the compiled BiomniBench-DA Public 50 comparison, earning **79.05** with **gpt-5.6-sol (xhigh)**. The result combines a Gemini 3.1 Pro judge score of **81.04** and a DeepSeek v4-pro judge score of **77.06** through an equal-weight mean, placing Open Science **#1** among the collected Public 50 results. Explore the [BiomniBench-DA dataset](https://huggingface.co/datasets/phylobio/BiomniBench-DA). ## Why Open Science Open Science brings research tasks, execution, files, and evidence into one local, inspectable desktop workspace. Research work is usually split across chat windows, notebooks, local scripts, scientific databases, file browsers, and reporting tools. Context is lost at every handoff, and the answer is often separated from the code and files that produced it. Open Science brings those pieces into one inspectable desktop workspace: - **Work that persists.** Projects, sessions, drafts, files, previews, and run history survive application restarts. - **Execution, not just suggestions.** The agent can run commands, Python