What is the Desktop GUI?

The Desktop GUI is a third-party community wrapper that runs the open-source deepseek-ai/deepseek-harness locally with a one-click app. It is a local GUI wrapper and desktop app: your API key stays local, no terminal needed, and every run is fully traceable. It is not affiliated with DeepSeek, unsigned and not notarized, experimental macOS Apple Silicon (arm64) only, requires macOS 13+.

What exactly is the deepseek-harness.io waitlist for?

The live experimental Desktop GUI is available now as an unsigned macOS arm64 build (requires macOS 13+). The waitlist is for community updates and feedback only — joining does not affect your ability to download the live build. This page is not affiliated with DeepSeek, and the app is a third-party community wrapper that runs locally with your own API key.

Live experimental · DeepSeek Harness GUI · macOS arm64

DeepSeek Harness GUI

One-click local wrapper · Your key stays local · No terminal needed · Fully traceable

DeepSeek Harness GUI is a live experimental Desktop App wrapping the open-source harness for local execution — available now for macOS Apple Silicon (arm64). No terminal or environment setup. Bring your own API key from DeepSeek / Anthropic / OpenAI. Your key stays local — third-party wrapper, not affiliated with DeepSeek, unsigned and not notarized.

Download for macOS — Apple Silicon v0.1.0 · arm64

SHA256 verifiedBUILD-INFONo Node requiredMIT · Unsigned

Explore Creator Edition →

Live experimental build: Download above is the GitHub Release dsh-desktop-v0.1.0 — Unsigned / Not notarized — Experimental · macOS Apple Silicon (arm64) only · Requires macOS 13+. Verify SHA256, then Gatekeeper requires manual open: right-click → Open or xattr -cr "DeepSeek Harness Desktop.app" (at your own risk). Third-party wrapper around the official open source deepseek-ai/deepseek-harness, maintained by an independent community, not affiliated with DeepSeek.

DeepSeek Harness upstream interface — Providers and Models list, MIT-licensed — Upstream interface · MIT-licensed

Live experimental build available above — the list below is for community updates and feedback only.

Local by design — code and credentials stay on your machine; the app uses your own model key and never uploads it.

Unofficial community waitlist — not affiliated with DeepSeek · independent community. Joining costs nothing and does not charge your card. MIT-licensed core: github.com/deepseek-ai/deepseek-harness

What it is

What is DeepSeek Harness?

It is an open-source agent harness from DeepSeek AI — a v0.1 developer preview under the MIT license. A harness is the runtime that turns a language model into a working agent: tools such as file editing, shell and search, session management, permission policies, and a record of every step so a run can be inspected, resumed or replayed.

Its design principle is that everything is a plugin — models, tools, skills, sessions, sandboxes, storage, loops, scheduling and the UI — on the Cordis framework. It ships four presets (Standard, Code, Minimal, Creator) and writes an append-only session log powering a trajectory view of prompts, reasoning, tool calls and results.

The official way to run it today is a local command, npx @deepseek-ai/dsh web, which assumes Node.js, a terminal and a configured model provider. You bring your own key from DeepSeek, Anthropic, OpenAI or any OpenAI-compatible endpoint — the model adapter is a plugin, so switching providers needs no framework changes. The Desktop app aims to handle installation, setup and updates, so the open-source engine stays the same underneath.

Want to see how it works end to end? Follow our DeepSeek Harness tutorial, or go straight to the install guide if you are ready to try the official command-line version today.

Download · Honesty note · Trajectory Ruler

DSH Desktop Download — macOS arm64

A third-party desktop wrapper around the official open source deepseek-ai/deepseek-harness. Your key stays local and never passes through our servers. The download appears only after a reproducible build passes verification.

Version info · Build honesty

Version: v0.1.0 Building
System: macOS Apple Silicon (arm64)
Size: 468M · GitHub Release · arm64 only · Requires macOS 13+
SHA256: 118168931c3e3064b0a785e35011f08f0c4100c7c923ab1671933266f8886439 · 11816893… · GitHub Release .sha256 BUILD-INFO Release
Entry: DeepSeek Harness Desktop.app / Executable after extraction
License: MIT · Third-party attribution consistent with upstream

Verify (GitHub Release)

shasum -a 256 -c dsh-desktop-0.1.0-mac-arm64.zip.sha256
cat BUILD-INFO.txt
# Local reproducible mirror: shasum -a 256 -c artifacts/internal/dsh-desktop-0.1.0-mac-arm64.zip.sha256

Live experimental build: GitHub Release dsh-desktop-v0.1.0 (468M, SHA 118168931c3e3064b0a785e35011f08f0c4100c7c923ab1671933266f8886439, arm64) is available now as an unsigned, not-notarized asset. Gatekeeper will block it — right-click → Open or xattr -cr "DeepSeek Harness Desktop.app" (at your own risk). Local reproducible build remains at artifacts/internal/dsh-desktop-0.1.0-mac-arm64.zip; Cloudflare releases/ stays honest-pending until signed. Verify: shasum -a 256 -c dsh-desktop-0.1.0-mac-arm64.zip.sha256 and fallback shasum -a 256 -c artifacts/internal/dsh-desktop-0.1.0-mac-arm64.zip.sha256. BUILD-INFO at https://github.com/18815292408/deepseek-harness-desktop/releases/download/dsh-desktop-v0.1.0/BUILD-INFO.txt and artifacts/internal/BUILD-INFO.txt. This page is an independent community, not affiliated with DeepSeek.

Trajectory ruler — zoomed

Visualizing the append-only session log — each step traceable

The ruler is the log's scale: a 1px line marked in execution order — prompt start, reasoning segment, three tool_call notches, permission gate diamond, result fill. Hover a notch to see the tool name.

Tool call shapes: file_edit square · shell slanted · search round · gate diamond. Only line and notches, no gradients/shadows/decorative orbs.

Open Source · GitHub trusted entry

Community-published, verifiable third-party wrapper source and experimental build artifacts. The primary live download is the GitHub Release (external, unsigned experimental) — see the button above; Cloudflare releases/ remains honest-pending until signed. The entries below are trusted GitHub links.

Open Source · github.com/18815292408/deepseek-harness-desktop Experimental Release · dsh-desktop-v0.1.0 — Unsigned, Not notarized

Boundary: Unsigned / Not notarized — Experimental · macOS Apple Silicon (arm64) only · Requires macOS 13+ · Not signed, not notarized — Gatekeeper requires Right-click → Open or xattr -cr (at your own risk) · deepseek-harness.io is an independent community page, not affiliated with DeepSeek · MIT license (wrapper and upstream both MIT) · No claim of signed/notarized/certified or official affiliation.

Repository and Release asset verification (experimental, arm64 only):
gh repo view 18815292408/deepseek-harness-desktop --json visibility,nameWithOwner · shasum -a 256 -c dsh-desktop-0.1.0-mac-arm64.zip.sha256 · SHA256 118168931c3e3064b0a785e35011f08f0c4100c7c923ab1671933266f8886439 (prefix 11816893) matches the internal build, see repository README and Release notes.

FAQ

Frequently asked questions

What is a coding harness?

A coding harness is the runtime around a language model that makes it a working agent: tool execution (file editing, shell, search), session management, permission policies and a traceable log. DeepSeek Harness is an open-source harness where every one of those capabilities is a replaceable plugin.

What exactly is this waitlist for?

The live experimental Desktop GUI is available now for macOS Apple Silicon (arm64, requires macOS 13+). The waitlist is for community updates and feedback only — joining reserves no exclusive access, costs nothing, and does not affect downloading the live experimental build. It helps us prioritize fixes and reach you with updates.

Is DeepSeek Harness free to use?

The source code is free under the MIT license. Running an agent still needs model access: you configure a provider such as the DeepSeek API with your own key, and tokens are billed by that provider. The desktop app does not change that — you keep using your own key.

Can it run models other than DeepSeek's?

Yes. The official documentation covers other providers and custom OpenAI-compatible endpoints. Because the model adapter is itself a plugin, swapping models does not require changing the framework source.

Is the harness ready for production use?

Not yet for production. DeepSeek labels the harness a developer preview and warns that breaking changes are coming. The project changes quickly, so anything built on top needs to track changes deliberately. The live experimental Desktop GUI above is available now for testing, and the community list helps us gather feedback while we iterate.

Do I need my own API key?

Yes. You add your own key in Settings → Models, and tokens are billed by that provider. The Desktop app keeps that model: your key stays on your machine and is never sent to us.

How do I get started today?

Install Node.js, run npx @deepseek-ai/dsh web, then open the URL it prints, add your key under Settings → Models and pick a workspace. Our install guide covers each step, and the tutorial explains the concepts.