BioDSH

agent
Guvenlik Denetimi
Gecti
Health Gecti
  • License — License: MIT
  • Description — Repository has a description
  • Active repo — Last push 0 days ago
  • Community trust — 63 GitHub stars
Code Gecti
  • Code scan — Scanned 12 files during light audit, no dangerous patterns found
Permissions Gecti
  • Permissions — No dangerous permissions requested

Bu listing icin henuz AI raporu yok.

SUMMARY

The bioinformatics agent desktop for clinicians and wet-lab scientists — built on DeepSeek Harness. One-click installers, a skill store, offline mode.

README.md

BioDSH — A workspace for biomedical research

The bioinformatics agent desktop for clinicians and wet‑lab scientists — built on DeepSeek Harness.

Everything is a plugin. Every analysis is a sentence.

中文说明 · Website · Download · DeepSeek Harness


BioDSH is a one‑click desktop app that puts a bioinformatics agent in front of people who don't use a terminal. Drop in your data, say what you want in plain language, and read the conclusion. It runs locally, never modifies your original files, and can run fully offline inside a hospital or lab intranet.

It is a thin, friendly shell around DeepSeek Harness (dsh): the agent runtime, the plugin system, and the reproducible session log are all dsh. BioDSH adds a clinician‑facing UI, a bioinformatics skill store, a bundled Python analysis environment, and a set of evaluated official skills.

The Sky interface · Design preview

The Sky interface ships in the 1.0.0 Preview (Windows). The screenshot below is a design render: catalog entries, counts and states are illustrative, not analysis results. The stable channel and auto‑update stay on 0.3.0 until 1.0 is final.

BioDSH Skill Hub — interface design preview, illustrative content

Interactive showroom · Brand & UI assets · Download the full kit

Download

Try 1.0.0 Preview — the new Sky interface, project/task tree, skill hub, pipelines and model sources, on DeepSeek Harness 0.2.1. Windows installer: BioDSH 1.0.0 Preview 1. It is a pre‑release: 0.3.0 users are not auto‑updated to it, and it installs over 0.3.0 in place (back up important projects first). Please report problems in Issues.

Stable installers are in the latest release:

Platform File Notes
Windows 10/11 BioDSH_*_x64-setup.exe Double‑click to install. On first run click More info → Run anyway.
macOS 12+ (Apple silicon) BioDSH_*_aarch64.dmg Unsigned — see the macOS note below.
Linux No 0.3.0 installer See build-from-source instructions below.

macOS: BioDSH isn't signed with an Apple Developer certificate, so macOS may say the app is “damaged and can’t be opened.” It isn’t — that’s Gatekeeper blocking unsigned apps. Drag BioDSH to Applications, then in Terminal run:

sudo xattr -cr /Applications/BioDSH.app

Enter your login password (it stays hidden) and open BioDSH again. You only need to do this once per install.

The app checks for new releases in the background and offers a one‑click download → verify signature → install → restart. Nothing is sent anywhere; the update feed is just this repo's latest.json.

What it does

  • Everything is a plugin. The best open‑source bioinformatics skills — 2,000+ community skills organised into 10 domains — plus a handful of evaluated, offline, reproducible official skills. Install one with a click.
  • Every analysis is traceable. Everything the model sees goes into an append‑only session log. In the chat, intermediate steps fold into a single line and expand on demand; conversations export for an advisor or reviewer.
  • Four real example projects ship with the app: single‑cell analysis & plots, literature review, public‑database retrieval, and desktop control with Zotero.
  • No coding. A one‑click Python environment (scanpy, anndata, pandas, matplotlib) is bundled; skills that need it install it once.
  • Desktop control. The agent can drive other apps (Excel, Prism, SPSS…) with real mouse and keyboard, with a clear "BioDSH is controlling the computer" banner on screen.
  • Data safety & offline. Original files are never modified; each analysis writes into a new subfolder. Fully‑offline mode talks to a local or intranet model and makes no outbound requests.

How it works

┌────────────────────────────────────────────┐
│  BioDSH (Tauri desktop shell)               │
│  • clinician‑facing UI (React)              │
│  • skill store · data view · environment    │
│  • bundled Python + official skills         │
│  ┌──────────────────────────────────────┐  │
│  │  DeepSeek Harness (dsh)               │  │
│  │  agent runtime · plugins · session log │  │
│  └──────────────────────────────────────┘  │
└────────────────────────────────────────────┘

A "skill" is a dsh plugin: a folder with a SKILL.md (instructions the agent follows) and optional scripts. The official skills live in biodsh-core/skills/; the desktop app is in desktop/.

Use with vanilla DeepSeek Harness

As a dsh plugin (npm: dsh-biodsh)

npm

The whole BioDSH agent — persona, 17 validated skills, the 2,000+ skill library, automatic skill selection and the convergence guard — as one dsh bundle. It adds a BioDSH agent preset and leaves your other presets alone. Requires dsh 0.2.1+ and pnpm on PATH.

dsh plugin --profile web add dsh-biodsh

Or in the dsh web UI: Plugins → Install → dsh-biodsh. Restart dsh and pick BioDSH in the preset menu. Details: packages/dsh-biodsh.

Skills only

BioDSH's official skills are ordinary dsh-native SKILL.md skills, so plain dsh picks them up automatically — no BioDSH app required. dsh scans a few skill folders; the simplest is a per-project .agents/skills/:

git clone https://github.com/sagirimo/BioDSH
cd BioDSH
# copy the skills into your dsh project (or run it from inside that project for ./.agents/skills)
./scripts/install-into-dsh.sh /path/to/your/dsh-project/.agents/skills
#   --global  installs into ~/.agents/skills instead

On Windows use scripts/install-into-dsh.ps1. Start dsh from that workspace and the BioDSH skills are available. (The scRNA / plotting skills expect a Python env with scanpy, anndata, pandas and matplotlib on PATH.)

Build from source

Prerequisites: Node.js 22+, Rust (stable), and the platform Tauri prerequisites (on Linux: libwebkit2gtk-4.1-dev libappindicator3-dev librsvg2-dev patchelf).

cd desktop
npm ci                        # install dependencies
node scripts/stage-dsh.mjs    # stage the dsh runtime
node scripts/sync-skills.mjs  # official skills → resources + store catalog
npm run tauri dev             # run the app
# or a full installer build:
npm run tauri build

The release recipe is at docs/release.workflow.yml — copy it to .github/workflows/release.yml, then push a tag such as desktop-tauri-v0.2.1 to build installers for all platforms and publish a release with latest.json. Auto‑update signing needs two repository secrets, documented at the top of that file.

Repository layout

Path What
desktop/ The Tauri desktop app — React UI (src/), Rust backend (src-tauri/), build scripts (scripts/), store catalog (store/).
biodsh-core/skills/ The evaluated official skills shipped with the app.
packages/dsh-biodsh/ The dsh-biodsh npm bundle for vanilla dsh (BioDSH preset, skill selection, convergence guard).
desktop/resources/ Bundled resources: the analysis‑environment spec, example‑project sessions, helper scripts.

License

MIT for BioDSH's own code. Bundled and downloaded components — DeepSeek Harness, community skills, the Python toolchain and packages — keep their own licenses.

Acknowledgements

Built on DeepSeek Harness. BioDSH is an independent project and is not affiliated with or endorsed by DeepSeek.

Yorumlar (0)

Sonuc bulunamadi