MIT 授權、DeepSeek 原生程式設計代理，單一 Go 二進位檔發布，支援終端、桌面、瀏覽器與編輯器（ACP）四種入口；獨立執行，README 未描述與 DeepSeek Harness（dsh）的整合。
DSH 整合
生態系相關
作者聲明
安全稽核
未稽核
最後核實
2026-08-21
授權條款
MIT
01它能幫你完成什麼？
Run long autonomous coding tasks from the terminal with an agent you can leave running
Code changes produced by a single local engine, guarded by plan mode, permissions, a workspace sandbox and per-turn checkpoints so runs stay readable and undoable
Developers who want a DeepSeek-native coding agent for long unattended terminal sessions
Drive the same local Reasonix engine from your editor
Native chat, editor context, tool-call approvals, model selection and workspace sessions in VS Code / VSCodium / Eclipse Theia via the `reasonix acp` backend
Developers who prefer coding-agent sessions inside their editor instead of a separate terminal
02如何將外掛接入 DeepSeek Harness？
先決條件
- npm on any supported platform, or Homebrew on macOS, for the prebuilt native binary path
- Go 1.25+ only if building the CLI from source (desktop builds additionally need Node 24+, pnpm 10 and the Wails CLI)
安裝步驟
- 01
Install the CLI: `npm i -g reasonix` (any OS; pulls the prebuilt native binary) or `brew install esengine/reasonix/reasonix` (macOS)
$ npm i -g reasonix
- 02
Run `reasonix setup` to configure a provider and model
- 03
Start an interactive session with `reasonix`, or run a one-off task with `reasonix run "implement the TODOs in main.go"`
驗證整合成功
作者未說明
03DSH 整合程度與能力邊界
Standalone DeepSeek-native coding agent — the README describes no DeepSeek Harness (dsh) integration; ecosystem-level relation via the built-in DeepSeek model preset
Autonomous coding runs across four entry points
a task or prompt from the terminal CLI/TUI, desktop app, browser, or an editor over ACP→code changes from one local engine, with plan mode, permissions, a workspace sandbox and per-turn checkpoints
the agent writes and modifies project files in your workspace during a runmodel calls go to the configured provider endpoint over the networkConfig-driven multi-model engine
a `reasonix.toml` declaring providers, the agent, enabled tools and plugins→no hardcoded models — DeepSeek as a preset, any OpenAI-compatible endpoint as a config entry, optional executor + planner pair in separate cache-stable sessions
MCP servers and Extension Protocol plugins
MCP servers or Extension Protocol v1 sidecars declared in the config→extra tools, prompts and resources from MCP; sidecars can intercept runtime events, contribute Providers and structured UI, and ship versioned plugin packages
extension sidecars can intercept runtime eventsCache-aware context maintenance
a long-running session against prefix-cache-friendly endpoints→a small stable environment summary injected at startup; stale tool output snipped/pruned before summary compaction
04適合誰？何時不該用？
適合
- Developers who want a DeepSeek-native coding agent for long unattended terminal sessions
- Developers who prefer coding-agent sessions inside their editor instead of a separate terminal
不適合
- Reasonix does not work out of the box without a model provider: you must run `reasonix setup` to configure a provider and model before starting a session, and all generation traffic goes to that configured endpoint.
05相容性、維護與安全提醒
- Reasonix does not work out of the box without a model provider: you must run `reasonix setup` to configure a provider and model before starting a session, and all generation traffic goes to that configured endpoint.
- Prebuilt binaries cover darwin/linux/windows × amd64/arm64, but building from source requires Go 1.25+ (pinned toolchain), and the desktop build additionally needs Node 24+, pnpm 10 and the Wails CLI.
- The README describes no DeepSeek Harness (dsh) integration: Reasonix is a standalone agent, so there is no documented in-harness install path — its DeepSeek relation is that DeepSeek ships as a built-in model preset.
MIT · actively maintained (latest release desktop-v1.31.0, 2026-08-20)
06常見問題
如何安裝 DeepSeek Reasonix？
執行 `npm i -g reasonix`（任何作業系統，自動取得預先建置的原生二進位檔），或在 macOS 使用 `brew install esengine/reasonix/reasonix`，接著執行 `reasonix setup` 設定供應商與模型。每個 GitHub release 都附 darwin/linux/windows × amd64/arm64 預編包與 SHA256SUMS。
它能與 DeepSeek Harness（dsh）整合嗎？
README 未描述任何 dsh 整合。Reasonix 是獨立執行的代理，它與 DeepSeek 生態的關聯在於：DeepSeek 以內建模組預設提供，任何 OpenAI 相容端點也可作為設定項目寫入 reasonix.toml。
可以在 VS Code 裡使用嗎？
可以——先安裝 CLI，再從 Visual Studio Marketplace 或 Open VSX Registry 安裝擴充功能 `SivanLiu.reasonix-agent`。它會啟動本機的 `reasonix acp` 後端，提供原生對話、編輯器脈絡、工具呼叫核准、模型選擇與工作區工作階段。
模型與工具如何設定？
供應商、代理、啟用的工具與外掛全部宣告在 `reasonix.toml` 中，沒有硬編碼模型。也可同時執行兩個模型（執行者＋規畫者），各自使用快取穩定的工作階段；MCP 伺服器可提供工具、提示與資源。
長時間自主執行的安全性靠什麼保障？
計畫模式、權限控管、工作區沙箱與逐輪檢查點，讓長時間自主執行始終保持可讀、可復原。整個引擎是自包含的靜態二進位檔，目標機器上除了二進位檔本身不需安裝任何東西。
07相關的 DSH 工作流程
mnemon
作者 mnemon-dev
LLM 監督的持久記憶系統，基於圖結構召回、跨會話知識共享，單個二進位制檔案，相容 DeepSeek Harness 等執行時。
phi
作者 pulseaiclub
來自 pi 的編碼智慧體，支援無限提供方、子智慧體、行內編輯與許可權門控。
sivtr
作者 ariestar
A unified agent memory workspace for human and agent | 一個統一的agent記憶工作空間
baro
作者 jigjoy-ai
把目標自動推進為 Pull Request 的命令列外掛，並提供 Mozaik 執行時沙箱，用於測試併發 AI 程式設計智慧體。
08資料與來源
DeepSeek ships as a preset; any OpenAI-compatible endpoint is a config entry, not new code.
One local engine, four ways in — terminal, desktop app, browser, or your editor over ACP. Plan mode, permissions, a work…
此頁面根據專案公開文件、儲存庫中繼資料與 DSH Plugins 的結構化解析所產生；最後核實於 2026-08-21。發現錯誤？提交更正。
最佳 DeepSeek Harness 外掛
從全目錄挑出的 12 個值得優先安裝的外掛，涵蓋各個分類。