Product

Monaco IDE for DeepSeek Harness

AI Code Write is a Cursor-style Windows desktop: Monaco editor, agent chat, git, plugins, and a loopback-only Harness runtime — with installer and auto-updates.

Editor

Monaco editor + Harness agent

Open workspaces, edit in Monaco, and chat with the agent in the same window — explorer, terminal, git, and Harness tools without a browser tab.

  • Branded Electron shell with installer and auto-update
  • Loopback-only Harness runtime (127.0.0.1:3080)
  • Plugin Market, thinking levels, and hindsight memory by default

Agent

A harness that can actually change code

DeepSeek Harness runs on loopback only. The agent reads the workspace, edits files, searches, and shells out — you approve privileged tools before they run.

Agent — edit + tools Ask — answers only Plan — design first Debug — hunt failures Multitask — queue prompts

Terminal & git

Run commands, review the diff

A local PTY sits under the editor. Commit message and button sit above the graph. Source control lists git status and inline diffs before you keep a change.

  • Ctrl+` toggles the terminal
  • Commit above graph · branch picker · stash
  • Agent shell tools run without flashing console windows

Diffs

See what the agent changed

Working-tree diffs stay next to the explorer. Keep or discard the patch after you read it — the agent does not get a silent commit.

Keyboard first

Command palette for the whole app

Open Folder, Search in Files, Source Control, Models, Plugins, Check for Updates — type a few letters and hit Enter.

Ctrl+Shift+P Command palette
Ctrl+S Save
Ctrl+` Terminal
Ctrl+O Open folder
Ctrl+Enter Send to agent
Ctrl+, Settings

Plugins

Plugin Market on the activity bar

Search the catalog with instant cache and background refresh. Install Harness plugins into the local web profile — GitHub-hosted packages work when Git is installed.

Models & thinking

Pick a model and reasoning level

Chat model picker in the agent bar, Models panel for favorites and custom providers, and Thinking level in Settings (Auto / Off / Low / High / Max).

Chat model picker

Switch deepseek-v4-pro or Flash per session from the agent input bar.

Models panel

Favorites, custom providers, and apply as session default.

Thinking level

Auto schedules low/high/max from tool history via dsh-thinking-levels.

Settings

API key and thinking level in one place

Paste your DeepSeek API key once, choose Auto reasoning by default, and check for updates — all from Settings (Ctrl+,).

Local by design

Loopback only. Your machine, your files.

Harness is bound to 127.0.0.1:3080. There is no hosted agent cloud, no remote workspace, and no reason to expose the process to the LAN.

  • API key stored locally in app settings
  • Privileged edits and shell tools wait for approval
  • Auto-update from https://aicodewrite.nexgenus.cn/updates/

First run

  1. Install the Windows NSIS package
  2. Open Settings and paste your DeepSeek API key
  3. Open Folder on a project
  4. Chat in the Agent panel, or just start typing

Download the Windows build

Version 0.2.51 · x64 NSIS installer · auto-updates after that.

Get Windows license