DeepSeek Harness is an open-source agent harness from DeepSeek AI with plugin-first design, local Web UI, and developer preview for agent workflows.
DeepSeek Harness (dsh) is an open-source agent harness developed by DeepSeek AI. The README describes it as an architecture where "everything is a plugin" and says it is powered by Cordis, which is presented as the underlying composability framework.
The project is currently in developer preview, and the README explicitly warns that compatibility-breaking changes may still happen. Users can run it locally from npm with `npx @deepseek-ai/dsh web`, which starts a Web UI on `http://127.0.0.1:3080` by default, or build and run it from a repository checkout with pnpm.
The repository also includes detailed documentation and runnable examples covering architecture, development, APIs, workflows, and interface patterns. That makes the project useful for developers who want to experiment with agent harness concepts, understand the plugin model, or adapt the harness to custom tooling and runtime setups.
The project is organized around a plugin-first design, with the README stating that "everything is a plugin." This suggests the harness is meant to be extended through modular components rather than a closed set of fixed features.
The README documents a Web UI launch path and notes that it is served locally by default at `http://127.0.0.1:3080`. That gives users a browser-based way to interact with the harness after installation.
The repository includes a docs tree with architecture, development, tool catalog, API gateway, and workflow-related documents. This points to a substantial documentation set for understanding internals and building on the project.
The examples directory includes runnable demonstrations such as headless-agent, jsonrpc-agent, web-cordis, web-schedule, and mcp-memory. These examples help users explore different interface patterns and extension points without assembling everything from scratch.
The project references feedback through GitHub Discussions and a Discord community. That gives adopters a place to ask questions, report issues, and follow ongoing development.
Developers can use the harness to try the plugin-first runtime model and understand how different parts of the system fit together before building their own extensions.
Teams can launch the Web UI locally and use the harness as an interactive environment for testing agent behavior, workflows, and interface changes.
Builders can inspect the docs tree and examples directory to learn the architecture, development process, and extension points for creating custom integrations or runtimes.
Users who prefer hands-on samples can start from example projects such as headless-agent or jsonrpc-agent to see how non-interactive and machine-readable agent flows are structured.
Community contributors can report issues, discuss behavior changes, and follow the rapid iteration path through GitHub Discussions and the project’s community channels.
The repository README describes DeepSeek Harness as an open-source agent harness developed by DeepSeek AI. The project is in developer preview, so the authors warn that compatibility-breaking changes may still occur.
The README shows a quick-start path using Node.js and the command `npx @deepseek-ai/dsh web`, which starts the Web UI at `http://127.0.0.1:3080` by default. It also documents how to clone the repository and run it from source with pnpm.
The source material shows a Web UI, a plugin-oriented architecture, and example runtimes such as headless-agent, jsonrpc-agent, web-cordis, web-schedule, and mcp-memory. That makes it useful for experimenting with agent workflows and extension points rather than only a single fixed interface.
Yes. The README points users to GitHub Discussions for feedback and bug reports, and the repository also references contributing and development documentation.
The repository is public on GitHub and the README lists the license as MIT. The page content does not show product-specific commercial pricing for DeepSeek Harness itself.
CreateOS Sandbox is an isolated compute environment for running code and agent workloads in Firecracker micro-VMs with private networking and SDK, CLI, or MCP control.
ByteAsk is a terminal-first AI coding agent for C and C++ that edits repos and verifies changes with compilers, debuggers, sanitizers, and tests.
hob is an independent workspace for coding agents, with local control over sessions, terminals, history, routing, and follow-up work.
Manta AI is an autonomous web app testing tool that maps app behavior, catches regressions, and generates tests from a URL, no scripts or selectors needed.
SonOf connects to your repo and PM tool, audits your codebase, and turns approved work into shipped tickets with senior engineering review.
Ghost is a terminal-based AI assistant for chatting, code generation, and CLI tasks. Includes free models, supports Linux, macOS, Windows, and is open source.
