Install & run ↗
OS-specific setup, runtime requirements and the first local Web UI.
Source-verified · Version-aware · Independent
DeepSeek Harness,
step by step.
Install the official dsh launcher, configure models, review plugins and resolve real errors without confusing community projects for first-party software.
Developer previewMITRelease data checked Oct 4, 2026
historical test / linux-x64
$ node --version v22.22.1 $ npx @deepseek-ai/dsh@0.1.0-rc.7 web dsh web: HTTP 200 on an isolated local port # Publisher: @deepseek-ai # Status: release candidate # Test environment: no provider credentials
Historical rc.7 runtime recordHTTP 200
Launcher@deepseek-ai/dsh
npm latest tag0.2.0-rc.2
Python SDK0.1.5rc1
Default origin127.0.0.1:3080
Setup command builder
Build your first setup command.
Choose a surface and platform. The helper generates public setup commands only—it never asks for an API key.
Generated command
npx @deepseek-ai/dsh@0.2.0-rc.2 webHistorical 0.1.0-rc.7 test on Linux x64 with Node 22.22.1 and npm 10.9.4. Generated commands use the current npm tag; this is not a runtime test of that release.Read the full install guide →
Tools / New
Generate a provider patch without exposing a key.
Choose the official DeepSeek adapter or an OpenAI-compatible gateway and inspect the independently reviewed rc.7 settings patch. For current releases, follow the model settings and configuration-layer guides. This historical builder does not generate current profile patches.
Browser onlyNo key inputOpen config lab →
Choose a path
One guide for each layer.
Start with the base runtime. Add providers, workspaces and extensions only after each lower layer is verified.
Configure models ↗
API credentials, model routes and OpenAI-compatible endpoints.
Build with Python ↗
The official Python package, JSON-RPC example and safe workspace notes.
Extend with plugins ↗
Install, develop and review community extensions without skipping trust checks.
Resolve errors ↗
Issue-led diagnosis for Windows, Node, provider, MCP and session failures.
Understand the system ↗
Cordis layers, CLI profiles, privacy controls and product comparisons.
Why this guide exists
Official facts. Community reports, labeled.
The product is changing quickly. We pin every technical claim to first-party documentation, package metadata, a reproducible test or a clearly labeled official community report.
- 01First-party firstProduct page, repository, docs and official registries.
- 02Environment attached“Works” always names the version, OS and runtime.
- 03No secret collectionThe browser helper never asks for API keys or code.
Live issue map
Fix the exact failing layer.
01Fix the DeepSeek Harness Workspace Picker on WindowsOfficial Discussion #527→02Fix “AbortSignal.timeout Is Not a Function” in DeepSeek HarnessOfficial Discussion #311→03Fix crypto.randomUUID on DeepSeek Harness LAN HTTPOfficial Discussion #514 + secure-context requirement→04Fix DeepSeek Harness MCP Connection ErrorsOfficial dsh-v0.2.1-alpha.1 documentation review; historical runtime verification remains 0.1.0-rc.7 on Linux x64.→05Fix node-pty “pty.node” Missing in DeepSeek HarnessRepro: ignore-scripts install; normal install verified→
Start with the source
Never guess which package is official.
The official npm launcher is @deepseek-ai/dsh. The official Python package is deepseek-harness-sdk. Similar unscoped package names may be unrelated community software.