DSH Science

English | 中文

DSH Science is a community branch of DeepSeek Harness for reproducible Python and R work in durable agent sessions.

It follows the everything is a plugin architecture and adds a Science Session domain, a host-local runtime, and a staged path to a complete Science workspace.

Run

Run from source

Install Node.js ^22.19 or >= 24 and pnpm 11, then run the repository from source:

git clone https://github.com/omdsh-dev/dsh-science.git
cd dsh-science
pnpm install
pnpm run build
pnpm dsh web

The command starts the Web profile and prints its local URL. Add a DeepSeek API key under Settings → Models, then start a session in the workspace from which you launched DSH.

Science foundation

Phase 2 is complete and provides the runtime foundation for Science Mode:

• Durable Science Session events, strict replay, invariants, and a client-safe projection.
• Observation of configured existing Conda prefixes for Python and R.
• Fresh-process execution with direct argv, an explicit clean environment, private per-session scratch, and full file-write confinement.
• Cancellation, timeout, process-tree quiescence, environment drift detection, and bounded output collection.

The Science Mode execution chain, Science subsystem reference, and Science Runtime package describe the implemented interfaces and ownership.

Roadmap

Phase	Plan
3	Ship the Science preset, reconstructable environment context, state queries, and Python/R tools.
4	Add immutable PNG chart versions and evidence-linked Outcome publication.
5	Add the Science view to the existing Details column.
6	Complete product composition, built-package checks, GUI replay, and source closure.

The Science MVP Decision Record owns the complete staged design.

Profiles and plugins

A profile is an ordered list of plugin bundles. From a source checkout, manage the Web profile with the same pnpm-backed plugin command used across DSH:

pnpm dsh plugin --profile web add <package>
pnpm dsh plugin --profile web remove <package>

Use the dsh-plugin topic when publishing a DSH plugin so it appears in the shared ecosystem.

Documentation

The Web UI guide, CLI reference, Python SDK, and examples cover the main ways to use and extend DSH.

Start development with the development guide and architecture documentation. Agents follow AGENTS.md.

Community

DSH Science is maintained in the omdsh-dev community and builds on DeepSeek Harness.

Open an issue for bugs, proposals, and implementation discussions.

Contributing

Read CONTRIBUTING.md before contributing.

License

BSD 3-Clause

Third-party dependencies and their licenses are disclosed in THIRD_PARTY_NOTICES.md.