← Home Features Download 中文

EdgeSidekick User Guide

A local multi-agent AI workbench. This guide covers packaging the Windows desktop exe, configuring your model, then workbench modes, chat, themes, Skills, and basic settings.

Package as exe Configure model Workbench vs Social Chat Themes Skills Settings

1. Package as exe (Windows)

Build once on an online Windows x64 machine. Copy the Setup installer to the target PC and install there — the target machine does not need Python or Node.js preinstalled.

Requirements

Windows x64
Node.js 18+
A usable host Python (the script prepares an embeddable runtime; a project .venv works too)
Access to npm / pip mirrors (defaults: Huawei Cloud, npmmirror, Tsinghua PyPI)

Build command

From the repository root in PowerShell or CMD:

.\scripts\build-windows.bat

Equivalent:

powershell -ExecutionPolicy Bypass -File .\scripts\build-windows.ps1

Optional flags

Flag	Effect
`-SkipRuntime`	Reuse existing `packaging/windows/payload/python`
`-SkipUi`	Reuse existing `ui/dist`
`-SkipPlaywright`	Skip Chromium (smaller build; agent browser tools unavailable)
`-Force`	Force re-download of the Python runtime

Example (skip Chromium, reuse UI):

.\scripts\build-windows.bat -SkipPlaywright -SkipUi

Outputs & install

Artifacts land in desktop/dist/:

EdgeSidekick-Setup-<version>-win-x64.exe — NSIS installer (recommended)
EdgeSidekick-<version>-win-x64.zip — portable zip

Run the Setup .exe (defaults to the current user directory; admin usually not required)
Launch EdgeSidekick from the Start menu or desktop shortcut
On first launch, pick a workspace folder

Config and sessions default to %APPDATA%\EdgeSidekick\. Unsigned builds may trigger SmartScreen — choose “More info” → “Run anyway”.

Kylin / Linux (optional)

On Linux you can build .deb / AppImage / tar.gz:

bash scripts/build-kylin.sh

See packaging/kylin/README.md in the repo.

2. Configure the model

After the app opens and you pick a workspace, configure a model before chatting:

Open Settings → Model
Set the API Base URL (OpenAI-compatible endpoint, or local Ollama, etc.)
Enter your API Key (for local Ollama, leave blank or use a placeholder per your provider)
Choose / enter model names; optionally configure separate main, sub-agent, and compress models to save tokens
Save, return to chat, and send a test message

API keys stay on this machine (installed app: %APPDATA%\EdgeSidekick\). History and Memory stay local by default; chat still needs reachability to your model API (public, private network, or localhost).

3. Workbench vs Social

Workbench

Built for engineering tasks: main/sub-agents, file edits, shell, browser sandbox, task panel. Best for coding, docs, and commands.

Social / Session

Conversation-first UI with session lists and DM-style chat, still backed by the same agents. Best for ongoing Q&A and topic threads.

Both share the same local runtime and model config.

4. Multi-agent chat

Send a message; replies stream over SSE.
The main agent can orchestrate sub-agents. Writes, shell, and Memory changes usually need human approval.
Attachments, stop, edit & resend, and optional file restore are supported.

5. Themes & language

Light clean for daytime focus.
Dark immersive for long night sessions.
Chinese / English UI; preferences can persist separately for the site and the app.

6. Skills

Drop packs under src/skills/ (or the tenant skills directory).
Invoke with /skill so tools load on demand.
Bundled examples cover writing, debugging, decks, and more.

7. Basic settings

Workspace: Settings → pick a local folder; switch among recent workspaces.
Browser sandbox: desktop BrowserView / Web Playwright; Select Mode + browser_* tools.
Memory / Tasks / MCP: append or rewrite MEMORY.md (with approval); track tasks; hook external tools via MCP.

Slash commands: /help · /new · /clear · /skills · /skill <name> · /memory · /history · /browser · /stop

See root README.md and packaging/windows/README.md for deeper technical notes.