Skip to content
Back to search

open-science

Open Science Desktop — local-first, model-agnostic AI research workbench for macOS, Windows & Linux. Open-source Claude Science desktop alternative built on Tauri + MCP + agent skills.

Safety scan

A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.

Checked today

Check it deeper on your machine

It has changed since its scan on 11 Nov 2026.

As a Claude Code plugin, install it and then ask Claude Code:

/plugin marketplace add keithmackay/mackayi/plugin install scanskillsafety@mackayiDo a deep scan of https://github.com/ai4s-research/open-science with scanskillsafety

Or from a terminal, clone it and run it:

git clone https://github.com/keithmackay/scanskillsafety-skill ~/.claude/skills/scanskillsafetynode ~/.claude/skills/scanskillsafety/cli.cjs --deep 'https://github.com/ai4s-research/open-science'

It checks the exact commit: every file, its packages against OSV (api.osv.dev), how it compares with this page's scan (findsafeskills.com receives the repository's address), and, if you agree, Claude reading every file with your own Claude subscription. It never runs the code.

Badge for your README
findsafeskills safety rating

Markdown

[![findsafeskills safety rating](/api/listings/ai4s-research--open-science/badge.svg)](/listings/ai4s-research--open-science)

HTML

<a href="/listings/ai4s-research--open-science"><img src="/api/listings/ai4s-research--open-science/badge.svg" alt="findsafeskills safety rating" /></a>

Install

i

Install instructions from the README

As written by the repo's author. findsafeskills has not verified these steps.

Download the latest installer from the [Releases page](https://github.com/ai4s-research/open-science/releases/latest).

- **macOS**: `.dmg` / `.app`, Apple Silicon and Intel, macOS 13 Ventura or later. - **Windows**: NSIS `.exe`, Windows 10/11 x64 — installs per user, no admin needed. A `.msi` is also published for IT-managed deployment; pick one format and stay on it. - **Linux**: `.deb` and `.rpm` on x86_64 Linux.

The macOS packages are Developer ID signed, notarized, and stapled, so they open normally — no `xattr` workaround needed. Windows and Linux builds are not signed yet.

**Windows**: if SmartScreen appears, choose **More info -> Run anyway**.

**Linux**:

sudo apt install ./Open.Science_*.deb
# or
sudo rpm -i Open.Science-*.rpm

What was scanned

The files the safety scan read, as written by the repository's author and shown as plain text. Lines with a finding are marked.

README.md · README · 532 lines

View this file on GitHub (read at the repository's current version: the scan did not record which commit it read)

This file has changed since it was scanned, so this is not exactly the text that was scanned, and no line is marked. The findings are listed with the safety scan above.

  1. <div align="center">
  2. [![Open Science Desktop — Local-first AI research workbench](./docs/assets/banner.webp)](https://github.com/ai4s-research/open-science)
  3. # Open Science Desktop
  4. **Local-first, model-agnostic AI research workbench for macOS, Windows & Linux.**
  5. Formerly Open Science. An open-source desktop alternative to Claude Science and
  6. similar AI-for-science workbenches — built with Tauri, MCP, agent skills, and
  7. reproducible artifacts. It connects agents, notebooks, files, figures, reports,
  8. runs, and review into one auditable desktop workflow.
  9. <p>
  10. <b>English</b> ·
  11. <a href="./README.zh.md">简体中文</a> ·
  12. <a href="./README.ja.md">日本語</a> ·
  13. <a href="./README.es.md">Español</a> ·
  14. <a href="./README.de.md">Deutsch</a> ·
  15. <a href="./README.fr.md">Français</a> ·
  16. <a href="./README.ko.md">한국어</a>
  17. </p>
  18. <p>
  19. <a href="./LICENSE"><img src="https://img.shields.io/badge/License-MIT-blue.svg" alt="License: MIT"></a>
  20. <a href="https://doi.org/10.5281/zenodo.21351225"><img src="https://img.shields.io/badge/DOI-10.5281%2Fzenodo.21351225-1682D4" alt="DOI"></a>
  21. <a href="https://internscience.github.io/ResearchClawBench-Home/"><img src="https://img.shields.io/badge/%F0%9F%8F%86%20%231-ResearchClawBench-FFB300" alt="#1 on ResearchClawBench"></a>
  22. <img src="https://img.shields.io/badge/platform-macOS%20%7C%20Windows%20%7C%20Linux-blue" alt="Platforms">
  23. <img src="https://img.shields.io/badge/i18n-7%20languages-5B8DEF" alt="7 interface languages">
  24. <img src="https://img.shields.io/badge/built%20with-Tauri%202%20%2B%20React-24C8DB" alt="Built with Tauri + React">
  25. <img src="https://img.shields.io/badge/runtime-OpenCode-success" alt="OpenCode runtime">
  26. <a href="https://discord.gg/fWNMDKcd5P"><img src="https://img.shields.io/badge/Join-Discord-5865F2" alt="Join Discord"></a>
  27. <a href="http://makeapullrequest.com"><img src="https://img.shields.io/badge/PRs-welcome-brightgreen.svg" alt="PRs Welcome"></a>
  28. <a href="https://linux.do"><img src="https://img.shields.io/badge/Join-linux.do-orange" alt="linux.do"></a>
  29. </p>
  30. </div>
  31. ---
  32. ## News
  33. - **2026-11-16** — 🛟 **Steadier turns on slow machines and networks.** A busy runtime keeps its running turns instead of being restarted under them; a turn whose runtime died stops spinning; a send with photos gets time to land and goes out once; a failed message can be retried or edited; Kimi's coding plan goes direct; browser control works again. Plus fixes for terminals, file history and Windows. *(v0.7.3)*
  34. - **2026-11-07** — 🧠 **OpenCode 2, and more from every agent.** The bundled runtime moves to OpenCode 2, with your OpenCode 1 history backed up first; a model's reasoning shows in the thread; ACP agents bring their own slash commands, context usage and task list; DeepSeek Harness joins the catalog. Plus sharper CJK text in terminals and a window that keeps its size. *(v0.7.2)*
  35. - **2026-10-29** — 🖥️ **Terminals that stay, agents side by side.** Terminals now run outside the app and survive updates, relaunches and crashes; Claude Code, Codex and pi run as conversations beside OpenCode, and a session moves between terminal and conversation in one click. Plus several accounts per agent, theme variants, new models on slow networks, and Chinese punctuation that goes in on the first press. *(v0.7.1)*
  36. - **2026-10-26** — 🌐 **The workbench from any browser, and agents that talk to each other.** Splits, Screens and terminals now work over Remote Access, on a phone too, with name-and-password accounts; define your own Specialists with their own instructions, skills, connectors and model; and agents read and message each other's conversations and terminals through `osd`. Plus queued messages while the agent works, and a no-root Linux install. *(v0.7.0)*
  37. - **2026-10-04** — 🖱️ **Computer Use, terminals and editors.** The agent can now operate desktop apps like ImageJ, Origin and instrument software; panes hold terminals, files and a code editor, where Claude Code / Codex resume after a relaunch; plus find (⌘F), jump to any session or project (⌘J), and Linux ARM64 for DGX Spark. *(v0.6.0)*
  38. - **2026-08-18** — 🖥️ **Runs without a screen, and the terminal command comes with it.** `osd server` starts the whole workbench — workspace, agent runtime, and the *same* web UI — on a machine with no display, and `osd session send … --wait` drives it from a script or another agent. `osd` ships inside the desktop installer and puts itself on your PATH on first launch; on a server the archive needs nothing installed. Models, keys and approvals are all configurable from the terminal (`osd model`, `osd auth`, `osd approval`).
  39. - **2026-08-13** — 🔌 **Speaks the Agent Client Protocol, both directions.** Drive Codex, Gemini CLI, Claude Code, or any other ACP agent from inside this app — with its own models, history, and your MCP connectors — or drive Open Science itself from Zed, JetBrains, or Neovim. *(v0.4.0)*
  40. - **2026-08-01** — 🗂️ **Projects, memory, and full history.** Group sessions into named projects (import an existing repo *in place*, no copying), give the agent persistent global and project memory, and reach every past conversation through a searchable history with archive, restore, and export. *(v0.3.1)*
  41. - **2026-07-24** — 🪟 **Split-pane tiling.** Tile sessions side by side, drag panes to re-dock them, keep several independent Screens, and run a different model in each pane. *(v0.3.0)*
  42. - **2026-07-21** — 🌐 **Access from anywhere — even your phone.** A token-authenticated gateway serves the *real* desktop UI to a CLI, a browser on your LAN, or your phone (loopback by default; LAN is opt-in). Start a run at your desk and read the finished figure and report on your phone. *(v0.2.3)*
  43. - **2026-07-21** — 🧭 **Browser control.** The agent can drive your own Chrome — profile and logins intact — to read the live web the way you do, or an isolated private browser on demand. *(v0.2.3)*
  44. - **2026-07-09** — 🎉 **#1 on ResearchClawBench.** Open Science Desktop ranks #1 by scored-task average on [ResearchClawBench](https://internscience.github.io/ResearchClawBench-Home/), an end-to-end benchmark for autonomous scientific research agents (Pass@1 leaderboard).
  45. ---
  46. ## Contents
  47. - [✨ What it does](#what-it-does)
  48. - [🎬 See it in action](#see-it-in-action)
  49. - [🧪 Current capabilities](#current-capabilities)
  50. - [🔌 Skills and connectors](#skills-and-connectors)
  51. - [📦 Install](#install)
  52. - [🖥️ Headless & CLI (`osd`)](#headless--cli-osd)
  53. - [🚀 Build from source](#build-from-source)
  54. - [🔒 Safety and privacy](#safety-and-privacy)
  55. - [🗂️ Repository layout](#repository-layout)
  56. - [📌 Status](#status)
  57. - [🤝 Contributing](#contributing)
  58. - [📖 Citation](#citation)
  59. - [⚖️ License](#license)
  60. ## What it does
  61. **Runs the whole research loop** — from a broad direction to a finished paper:
  62. exploration, literature survey, hypothesis, experiment code, analysis, figures, and
  63. write-up, in one continuous, auditable session.
  64. - **Autonomous research agents** — the bundled `ai4s-agent` chains specialist skills
  65. end to end (explore → survey → experiment → write), and each stage drops a real,
  66. inspectable artifact into your workspace, not just a chat reply.
  67. - **Everything traces back** — figures, tables, reports, notebooks, and run outputs
  68. link to the exact code, inputs, environment, model output, and conversation that
  69. produced them.
  70. - **Local-first and yours** — sessions, data, provenance, notebooks, and run records
  71. live in local folders on your machine. Nothing leaves by default.
  72. - **Model-agnostic runtime** — the UI talks through `packages/sdk` to a bundled,
  73. pinned OpenCode sidecar. Bring your own model; providers, skills, and MCP servers
  74. stay pluggable.
  75. - **Reproducible by construction** — local, SSH/Slurm, Modal, and notebook-batch runs
  76. are captured as reproducible run records, not loose terminal scrollback.
  77. - **Reach it from anywhere** — a built-in, token-authenticated gateway serves the
  78. *real* desktop UI to a browser on your LAN or phone (or, with a tunnel, from
  79. anywhere) — kick off a run at your desk and check on it from your phone over lunch.
  80. Off by default; loopback-only until you opt in, and API keys never leave the machine.
  81. - **Drives your own browser** — the agent can control your real Chrome, with your
  82. profile and logins intact, to read the live web the way you do — or an isolated
  83. private browser when you'd rather it not.
  84. - **Plan before it acts** — `/plan` lays out an execution plan before touching a
  85. file, and `/goal` fixes the objective, constraints, and acceptance criteria the
  86. agent then works toward.
  87. - **Built for long projects** — named projects group their sessions, two layers of
  88. persistent memory (global and per-project) carry what matters between them, and a
  89. long conversation compacts itself as it approaches the model's context window.
  90. - **Work several threads at once** — tile panes side by side, keep independent
  91. Screens, and give each pane its own model.
  92. - **Extensible** — agent skills, MCP servers and one-click science connectors,
  93. `/` commands, `!` shell mode, and a model-agnostic SDK.
  94. ## See it in action
  95. **One prompt -> a publication-grade figure, and every point traces to the exact code
  96. and inputs that made it.** No black boxes: open any artifact to see its generating
  97. script, its data files, and the conversation that produced it.
  98. ![A rendered cross-species atlas figure beside its generating script and input files in the artifact inspector](./docs/assets/showcase-provenance.webp)
  99. **Literature -> a verifiable report.** Fan the search out across sources, draft a
  100. manuscript rendered as a PDF, and gate it on a citation review — DOIs resolved,
  101. unsourced numbers and figure/code inconsistencies flagged — before anything ships.
  102. ![A protein-language-model literature survey compiled into a PDF manuscript, with a citation reviewer confirming every DOI resolves](./docs/assets/showcase-literature.webp)
  103. **Drives your own Chrome.** The agent reads the live web through your real browser
  104. profile — logins and all — then turns what it finds into a figure and a sortable CSV.
  105. ![The agent driving the user's own Chrome via open-science-browser to harvest bioRxiv preprints into a chart and CSV](./docs/assets/showcase-browser.webp)
  106. **Research from anywhere — even your phone.** A built-in authenticated gateway serves
  107. the *real* desktop UI to a browser on your LAN (or a tunnel), so you can kick off a run
  108. at your desk and read the finished figure and report on your phone.
  109. <table align="center">
  110. <tr>
  111. <td align="center" width="33%"><img src="./docs/assets/showcase-mobile-home.webp" width="240" alt="The workbench in a phone browser: the new-session screen with starter analyses"><br><sub>New session</sub></td>
  112. <td align="center" width="33%"><img src="./docs/assets/showcase-mobile-run.webp" width="240" alt="A completed dose-response analysis — script, results, figure, and report — on a phone"><br><sub>A finished analysis</sub></td>
  113. <td align="center" width="33%"><img src="./docs/assets/showcase-mobile-reproduce.webp" width="240" alt="Reproducing an scVI benchmark, with its ARI-vs-epoch figure, viewed on a phone"><br><sub>A reproduced benchmark</sub></td>
  114. </tr>
  115. </table>
  116. <details>
  117. <summary><b>More screenshots</b></summary>
  118. <br>
  119. ![Reproducing an scVI integration benchmark on a remote A100 with a pinned environment, execution log, and provenance](./docs/assets/showcase-remote.webp)
  120. ![An 8-arm scVI hyperparameter sweep table beside a live analysis notebook sharing the agent's kernel](./docs/assets/showcase-experiment.webp)
  121. </details>
  122. ## Current capabilities
  123. **The research loop, as skills.** One meta-skill runs the full pipeline; each stage
  124. is a self-contained skill that produces a real, gradeable artifact — runnable on any
  125. model OpenCode supports:
  126. | Skill | Role | Primary output |
  127. | --- | --- | --- |
  128. | `ai4s-agent` | Runs the four skills below, in order | The full research package |
  129. | `research-explorer` | Turn a broad direction into concrete topics | `research_exploration.md`, `topic_matrix.md`, `literature_pre_survey.md` |
  130. | `literature-survey` | Write a literature survey | 6–20 pp PDF, 60+ real citations, LaTeX source, taxonomy figures |
  131. | `experiment-suite` | Build an experiment package | Design doc, runnable code, `results.json` with provenance, figures, report |
  132. | `paper-writer` | Write a research paper | 8–14 pp PDF, 200+ citations, 4–8 figures, tables |
  133. | `mindmap-render` | Render a mindmap | Image generated from a `topic_matrix.md` |
  134. | `integrity-auditor` | Audit a paper's integrity | Image / numerical / logical findings, 4-level evidence grading, `audit_report.md` |
  135. These ship in the `ai4s-skills` pack alongside first-party review skills and the
  136. office/document skills below.
  137. ### Platform
  138. | Area | Current state |
  139. | --- | --- |
  140. | Desktop shell | Tauri 2 + React + TypeScript + Vite, with macOS, Windows, and Linux desktop builds. |
  141. | Runtime | Bundled OpenCode sidecar, auto-started by the app, isolated from the user's own OpenCode config/data. |
  142. | Projects | Named project workspaces that group their sessions; import an existing folder in place (never copied) or adopt one already inside the workspace; move an existing session into a project. |
  143. | Sessions | Multi-session chat/history, dated workspace folders, searchable history with archive/restore/export, `@` file and `#` conversation references, `/` commands, and `!` shell mode. |
  144. | Layout | N-ary split-pane tiling with drag-to-dock, independent Screens, per-pane model and reasoning effort, and cross-screen pane drag. |
  145. | Agent modes | `/plan` for plan-then-execute, `/goal` for objective and acceptance criteria, live subagent status in its own panel, and Stop that reflects the runtime's real server state. |
  146. | Memory | Global and per-project memory layers, switchable, plus automatic context compaction as a conversation approaches the model's window. |
  147. | Remote compute | Register machines from your `~/.ssh/config`, probe them, and submit, track, or cancel jobs from the app. |
  148. | Appearance | Light, Warm, and Dark themes with per-theme accents, and UI zoom. |
  149. | Files | Global and per-session file browsing, context menu actions, external open/reveal, copy path, and local preview server. |
  150. | Headless & CLI | `osd server` runs the workbench with no window — same workspace, same runtime, same web UI, served from one self-contained directory — and `osd` drives it (or a running desktop app) from a terminal: sessions, projects, runs, files, approvals, `--wait`, `--json`. |
  151. | Remote access | Token-authenticated gateway that serves the real UI to a CLI, a LAN web browser, or your phone (loopback by default, LAN opt-in); read-only vs full access modes; copy a link with the token embedded to connect in one tap. API keys never cross the wire. |
  152. | Editor interop (ACP) | Speaks the Agent Client Protocol in both directions: run any ACP agent (Codex, Gemini CLI, Claude Code, …) as the runtime behind the ordinary UI, with its own model and reasoning selectors, history replay, and this app's MCP connectors; or let an external editor (Zed, JetBrains, Neovim, …) drive Open Science, reusing the gateway token. |
  153. | Browser control | The agent drives your own Chrome — profile and login state preserved — reading pages through the accessibility tree, or an isolated/private browser on demand. |
  154. | Computer use | The agent reads and drives apps on your own screen through the platform's accessibility API — native research software included — acting on named elements rather than guessing at pixels, and reporting whether each action could actually be verified. On macOS the Accessibility permission is held by a separate helper, not by the whole app; Screen Recording, which only optional screenshots need, is granted to Open Science itself because macOS attributes screen capture to the app that started the request. |
  155. | Notebooks | Real `.ipynb` files, Python and R notebook creation, local kernel execution, managed Jupyter environment via bundled `uv`, and an Open JupyterLab action. |
  156. | Runs | Append-only run logs, global SQLite run index, search/facets/pagination, local/remote surfaces, output links, logs, and reproduce prompts. |
  157. | Provenance | `.openscience/provenance.jsonl` tracks file versions and links produced artifacts back to the run or edit that created them. |
  158. | Review | Traceability, statistics-integrity, domain-check, large-file, publication-figure, remote-compute, and Modal run skills are bundled as first-party skills. |
  159. | Viewers | PDF, image, video, HTML, Markdown, code, CSV/TSV tables with charts, DOCX, XLSX, PPTX, molecules, 3D meshes, genome tracks, FITS, DOS/DOSCAR, EIGENVAL bands, qcode, anomaly maps, and phase files. |
  160. | Models | OpenCode provider catalog, OAuth/API-key provider flows, custom OpenAI-compatible endpoints, and local/provider-specific options supported by OpenCode. |
  161. | Interface languages | English, Simplified Chinese, Japanese, Spanish, German, French, and Korean. Portuguese (Brazil) and Arabic are registered but not selectable yet. |
  162. ## Skills and connectors
  163. Bundled skills are fetched for builds and releases instead of being committed into
  164. git history:
  165. - `ai4s-skills` pack from `ai4s-research/ai4s-skills`.
  166. - Office/document skills from the Apache-2.0 `anthropics/skills` repository:
  167. `docx`, `pdf`, `pptx`, and `xlsx`.
  168. - First-party core skills in `runtime/skills/core/`:
  169. `traceability-review`, `stats-integrity`, `domain-check`, `large-file`,
  170. `publication-figures`, `remote-compute`, and `modal-run`.
  171. One-click science MCP connectors currently include:
  172. - Literature search: arXiv, PubMed, Crossref, Semantic Scholar, bioRxiv/medRxiv.
  173. - Biomedical databases: PubMed, ClinicalTrials.gov, MyVariant/ClinVar.
  174. - Materials Project.
  175. - FRED economic data.
  176. - Space weather.
  177. - Open-Meteo weather and climate.
  178. - USGS water data.
  179. You can also add any local or remote MCP server from Settings. See
  180. [`docs/CONNECT_YOUR_TOOLS.md`](./docs/CONNECT_YOUR_TOOLS.md).
  181. For a neutral positioning note, see
  182. [`Open Science Desktop vs OpenScience`](./docs/open-science-desktop-vs-openscience.md).
  183. ## Install
  184. Download the latest installer from the
  185. [Releases page](https://github.com/ai4s-research/open-science/releases/latest).
  186. - **macOS**: `.dmg` / `.app`, Apple Silicon and Intel, macOS 13 Ventura or later.
  187. - **Windows**: NSIS `.exe`, Windows 10/11 x64 — installs per user, no admin needed. A `.msi` is also published for IT-managed deployment; pick one format and stay on it.
  188. - **Linux**: `.deb` and `.rpm` on x86_64 Linux.
  189. The macOS packages are Developer ID signed, notarized, and stapled, so they open
  190. normally — no `xattr` workaround needed. Windows and Linux builds are not signed yet.
  191. **Windows**: if SmartScreen appears, choose **More info -> Run anyway**.
  192. **Linux**:
  193. ```bash
  194. sudo apt install ./Open.Science_*.deb
  195. # or
  196. sudo rpm -i Open.Science-*.rpm
  197. ```
  198. No root? This installs everything under your home directory — `osd` by
  199. default, the desktop app with `--desktop` (which needs WebKitGTK 4.1 already on
  200. the system). `--uninstall` removes it again.
  201. ```bash
  202. curl -fsSL https://raw.githubusercontent.com/ai4s-research/open-science/master/scripts/install.sh | sh
  203. curl -fsSL https://raw.githubusercontent.com/ai4s-research/open-science/master/scripts/install.sh | sh -s -- --desktop
  204. ```
  205. ## Headless & CLI (`osd`)
  206. A research machine usually has no screen. `osd` is the same workbench without
  207. one: the same workspace layout, the same agent runtime, the same projects, and
  208. the same web UI — served over HTTP instead of drawn in a window.
  209. **On a server, take the archive.** `osd-<version>-<target>` from Releases
  210. unpacks and runs with nothing installed — verified on a bare Ubuntu container
  211. with no packages added at all.
  212. ```bash
  213. # Configure the machine (works before any server is running)
  214. ./osd auth set anthropic --key sk-… # stays on this machine, never on the wire
  215. ./osd model set anthropic/claude-opus-4-5 # the default for every turn
  216. ./osd server --lan # prints its URL and access token
  217. ```
  218. Keys never have to touch a file: the agent runtime inherits this process's
  219. environment, so `ANTHROPIC_API_KEY=sk-… ./osd server` needs no `auth set` at
  220. all. A self-hosted or proxied endpoint goes in the same command
  221. (`--base-url https://my-gateway.internal/v1`), and `osd auth ls` prints provider
  222. names only — no key is ever printed by anything. Changing a key needs a restart;
  223. the CLI says so rather than leaving you to wonder.
  224. Open the printed URL and you get the real desktop UI in a browser, phone
  225. included. Or drive it from a terminal — on the same machine, over SSH, or from
  226. your laptop:
  227. ```bash
  228. osd project new "Reef survey"
  229. id=$(osd session new --project "Reef survey")
  230. osd session send "$id" "Fit the 2015–2024 bleaching trend and write report.md" \
  231. --model anthropic/claude-sonnet-4-5 --wait
  232. osd fs ls figures/
  233. osd fs get report.md --output ./report.md
  234. ```
  235. On Windows the same commands work in PowerShell; only the shell's own syntax
  236. differs:
  237. ```powershell
  238. $id = osd session new --project "Reef survey"
  239. osd session send $id "Fit the 2015-2024 bleaching trend and write report.md" --wait
  240. ```
  241. **On your own machine it is already installed.** The desktop installer carries
  242. `osd`, and the app puts it on your PATH the first time it starts, so a new
  243. terminal has the command with nothing to set up. It writes one small wrapper
  244. (`~/.local/bin/osd`, or `~/bin` when a terminal already searches that) — never a
  245. symlink, because `osd` finds its runtime next to its real executable. If that
  246. folder is not on PATH, the app adds it to your login profile and Settings →
  247. Remote Access says which file it touched. Nothing else on your shell is changed.
  248. `--wait` returns when the turn is finished, not when it was accepted, and fails
  249. loudly if it produced no reply. `--json` prints the API's own response for
  250. scripts.
  251. ### Concurrency and per-project setup
  252. Parallel work does not need a second server, or a second workspace: one
  253. `osd server` already runs many sessions at once. The gateway is
  254. thread-per-connection and nothing on the session path holds a global lock —
  255. what keeps one session's work out of another's is the folder each one is pinned
  256. to. `osd session new --project NAME` runs a session in a per-project folder of
  257. the workspace (created with `osd project new NAME`), and
  258. `osd session new --directory DIR` pins one to any other folder.
  259. Per-project MCP follows the folder too: an MCP server declared in
  260. `<folder>/.opencode/opencode.json` applies only to sessions working in that
  261. folder, and every session shares the same sidecar. One caching rule makes the
  262. ordering matter: a folder's opencode configuration is read once, the first time
  263. the sidecar uses that folder, and cached from then on. Write that configuration
  264. before any session is pointed at the folder: `osd session new --project NAME`
  265. (or `--directory DIR`) already reads and caches it, before a single turn has
  266. run, so "before the first turn" is already too late. Changing it afterwards
  267. takes effect only after a server restart, and until then the session runs,
  268. reports success, and used the previous configuration.
  269. None of this needs a second installation of the workbench.
  270. ### Isolated state directories (--state-dir)
  271. `osd server --state-dir DIR` — or `OSD_STATE_DIR=DIR` — keeps everything one
  272. instance writes (session database, sidecar config, workspace record, logs)
  273. under DIR instead of the platform data directory. Nothing else changes: the
  274. default workspace still resolves to your real Documents folder, and the desktop
  275. app is unaffected, because it never goes through the headless override.
  276. This is NOT a way to get more concurrency — one server already runs many
  277. sessions at once (above). It is for genuinely separate installations on one
  278. machine: an unattended shard next to the desktop app, a benchmark cluster.
  279. Credentials stay per instance and are never copied: a fresh DIR has none, so
  280. the first server started there prints a note and the one-liner that adds
  281. one — `OSD_STATE_DIR=DIR osd auth set <provider> --key …` — or you export the
  282. provider's own variable (`ANTHROPIC_API_KEY=…`) before starting and the sidecar
  283. inherits it.
  284. ### Which model, and who approves what
  285. `osd model` shows the default, `osd model ls` lists what the runtime can
  286. actually serve (the providers this machine has credentials for, current one
  287. marked), and `osd model set <provider/model>` changes it — over the gateway, so
  288. it works against a remote server too. Any single turn can override it with
  289. `osd session send --model … --agent … --effort …`.
  290. Approvals still apply: the agent asks before running commands, deleting files,
  291. installing dependencies or reaching the network. Without a window, `--wait` names
  292. what is waiting and offers both answers — `osd permission ls` /
  293. `osd permission allow <id>` in the terminal, or the gateway URL it prints, which
  294. carries the token so a browser on your laptop or phone can approve it.
  295. For a machine with nobody watching, opt out explicitly:
  296. ```bash
  297. osd approval # what has to be asked today
  298. osd approval set full # never ask — commands, deletions, installs, network
  299. ```
  300. `full` is a deliberate choice, not a default: the agent stays confined to the
  301. workspace, but nothing pauses for you. `osd approval set approve` puts every
  302. rule back. `osd approval set full` rewrites the runtime's own config on this
  303. machine, so it has to run on the machine itself — over SSH, not through
  304. `--gateway`: the gateway deliberately refuses config writes, and says so rather
  305. than silently doing nothing.
  306. ### As a service
  307. `osd server` is an ordinary foreground process, so systemd runs it as-is. This
  308. unit was run end to end on Ubuntu — enable, restart, crash, stop:
  309. ```ini
  310. # /etc/systemd/system/osd.service
  311. [Unit]
  312. Description=Open Science Desktop (headless)
  313. After=network-online.target
  314. [Service]
  315. Type=simple
  316. User=ubuntu
  317. Environment=HOME=/home/ubuntu
  318. ExecStart=/opt/osd/osd server --port 4788
  319. Restart=on-failure
  320. RestartSec=3
  321. [Install]
  322. WantedBy=multi-user.target
  323. ```
  324. `sudo systemctl enable --now osd` and the printed URL and token land in
  325. `journalctl -u osd`. A unit is also the tidiest way to run it: systemd stops the
  326. whole cgroup, so the agent runtime never survives the server, however it dies.
  327. With no `--gateway` given, `osd` talks to a gateway already running on the same
  328. machine — including the desktop app's — so with the app open, `osd session ls`
  329. just works. Otherwise point it anywhere with `osd login --gateway <url> --token
  330. <token>`.
  331. What is *not* there without a desktop: local Jupyter kernels, native file
  332. dialogs, and the OS file manager — the web UI hides those rather than offering
  333. controls that would fail. Two more are worth knowing: **provenance and run
  334. records are written by the desktop client**, so a headless server keeps the
  335. workspace's file history through git snapshots but does not append to
  336. `provenance.jsonl` or the run index.
  337. ## Build from source
  338. Prerequisites:
  339. - Node.js >= 20
  340. - pnpm 9
  341. - Rust toolchain
  342. - macOS, Windows, or Linux system dependencies required by Tauri
  343. ```bash
  344. git clone https://github.com/ai4s-research/open-science
  345. cd open-science
  346. pnpm install
  347. # Fetch pinned sidecars and bundled skills. These are git-ignored.
  348. bash scripts/dev/fetch-opencode.sh
  349. bash scripts/dev/fetch-uv.sh
  350. bash scripts/dev/fetch-skills.sh
  351. # The osd terminal client is bundled too — it is ours, so it is built, not fetched.
  352. bash scripts/dev/build-osd-sidecar.sh $(rustc -vV | sed -n 's/host: //p')
  353. # Run in development or build installers.
  354. pnpm --filter @ai4s/desktop tauri dev
  355. pnpm --filter @ai4s/desktop tauri build
  356. ```
  357. Useful checks:
  358. ```bash
  359. pnpm test
  360. pnpm typecheck
  361. pnpm lint
  362. ```
  363. ## Safety and privacy
  364. - Workspace files, raw data, session history, provenance, notebooks, and run records
  365. stay local by default.
  366. - Command execution, file deletion, dependency installation, and remote connections
  367. are human-approved flows in the desktop app.
  368. - Provider credentials are written to app-private runtime config, not to the
  369. workspace, provenance, git, exports, or global OpenCode config.
  370. - Settings includes a plain-language data-flow view explaining what can be sent to
  371. the selected model provider.
  372. ## Repository layout
  373. | Path | Purpose |
  374. | --- | --- |
  375. | `apps/desktop/` | Tauri + React desktop app. |
  376. | `packages/sdk/` | `OpenCodeClient`; keeps the UI from calling OpenCode directly. |
  377. | `packages/shared/` | Shared domain types and chart palette. |
  378. | `packages/ui/` | Shared UI package. |
  379. | `runtime/skills/core/` | First-party scientific skills. |
  380. | `runtime/skills/external/` | Build-fetched external skills. |
  381. | `runtime/harness/` | Runtime harness knowledge and operator context. |
  382. | `runtime/mcp/` | MCP runtime notes/configuration. |
  383. | `examples/` | Built-in example workspaces. |
  384. | `crates/osd-core/` | The server core — workspace, sidecar, gateway. No Tauri, so it runs headless. |
  385. | `crates/osd-cli/` | `osd`: the headless server and its client. |
  386. | `scripts/dev/` | Sidecar, `uv`, skill fetchers, and focused regression probes. |
  387. | `docs/` | Product, technical, operator, connector, and research notes. |
  388. ## Status
  389. The project is a working desktop MVP in active development. The most reliable current
  390. implementation log is [`PROGRESS.md`](./PROGRESS.md). Product and architecture notes
  391. live in [`docs/PRD.md`](./docs/PRD.md) and
  392. [`docs/TECHNICAL_DESIGN.md`](./docs/TECHNICAL_DESIGN.md), but those documents include
  393. target design as well as historical status notes.
  394. Near-term work is focused on Windows code signing, auto-update, broader
  395. Windows/Linux verification, richer connector hardening, and continued
  396. reproducibility review. macOS releases are already signed and notarized.
  397. ## Contributing
  398. Issues and PRs are welcome. Keep changes minimal and verifiable, follow
  399. [`AGENTS.md`](./AGENTS.md), and run the checks before opening a PR. For discussion,
  400. join the [Open Science Discord](https://discord.gg/fWNMDKcd5P) or the
  401. [linux.do](https://linux.do) community.
  402. ## Citation
  403. If you use Open Science Desktop in your research, please cite it:
  404. ```bibtex
  405. @software{open_science_desktop,
  406. author = {{The Open Science Desktop Contributors}},
  407. title = {Open Science Desktop: a local-first, model-agnostic AI research workbench},
  408. year = {2026},
  409. version = {0.7.3},
  410. doi = {10.5281/zenodo.24286110},
  411. url = {https://github.com/ai4s-research/open-science},
  412. license = {MIT}
  413. }
  414. ```
  415. GitHub's **"Cite this repository"** button (top of the repo page, generated from
  416. [`CITATION.cff`](./CITATION.cff)) provides the same reference in APA and BibTeX.
  417. ## License
  418. [MIT](./LICENSE). Bundled third-party skills and connectors keep their own licenses.
  419. Computer use is a port of the design and native providers from [Orca](https://github.com/stablyai/orca) (MIT) — an accessibility-tree-first approach, with an explicit verified/unverified result on every action, that is simply the right way to build this. Our thanks to that team for doing it well and giving it away.
  420. > Open Science Desktop is beta research tooling. Treat outputs as drafts: verify numbers,
  421. > citations, code, and conclusions before publication or decision-making.

Ratings and comments

Sign in to rate this skill.

Sign in

Free to use. If it’s helped, pass it on.