Skip to content

DotCraft Getting Started

This path is for first-time DotCraft users: install Desktop, choose a project folder, configure a model provider, then run your first session. After that, move into TUI, AppServer, API, SDK, or Automations as needed.

Quick Start

1. Download Desktop

Go to GitHub Releases and download the desktop app:

PlatformRecommended file
WindowsDotCraft-Desktop-win-x64-Setup.exe
macOSDotCraft-Desktop-macos-x64.dmg

Desktop is the recommended first entry point because workspace selection, model configuration, sessions, diffs, plans, automation review, and runtime status live in one UI.

DotCraft Desktop

If you prefer building from source, install the .NET 10 SDK, Rust toolchain, and Node.js, then run this from the repository root:

bash
build.bat

On Linux / macOS, run:

bash
bash build_linux.bat

2. Initialize a Workspace

On first launch, choose a real project folder as the workspace. DotCraft keeps that project's configuration, sessions, tasks, skills, and attachments with the project, so Desktop, terminal, and automation entry points can continue from the same context.

To complete first-time setup from a terminal, run this in a project directory that does not yet have .craft/:

bash
dotcraft

Start from a real project folder instead of an empty directory so the agent can read repository structure, existing docs, and build scripts.

Workspace setup wizard

3. Configure a Model

DotCraft supports two common model paths:

PathBest for
OpenAI-compatible API keyOpenAI API, OpenRouter, DeepSeek, and compatible providers
CLIProxyAPIReusing a local coding-agent CLI through an OpenAI-compatible proxy

API proxy configuration

The minimal configuration usually looks like this:

json
{
  "ApiKey": "sk-your-api-key",
  "Model": "gpt-4o-mini",
  "EndPoint": "https://api.openai.com/v1"
}

Put sensitive values in global configuration, and put project-specific model, tool, and entry-point settings in the current workspace configuration. If you need to edit files directly, the paths are global ~/.craft/config.json and workspace <workspace>/.craft/config.json. See Configuration & Security for the full reference.

4. Run the First Session

Open the workspace in Desktop, create a session, and send a lightweight request:

text
Read this repository's README and docs/index.md, then tell me how to start the project.

If you prefer a script-friendly command-line entry, run a one-shot task from the project directory:

bash
dotcraft exec "Read this repository's README and docs/index.md, then tell me how to start the project."

In an initialized workspace, dotcraft does not enter an interactive chat. Use the TUI for terminal interaction.

For a richer terminal UI, continue with the TUI Guide.

Understand the Entry Model

DotCraft organizes its entry points around the Unified Session Core: command-line runs, Desktop, IDEs, bots, and automations do not each maintain their own agent loop, but reuse the same execution engine and session model.

DimensionGatewayUnified Session Core
Client customizationHard to customize once everything is flattened into a message busFlexible, native client experiences
Approval / HITLCannot express platform-native approval flowsRendered with native platform UI
Cross-channel resumeNot supportedConversations can resume across channels
Workspace persistenceNot supportedDesigned around the workspace

Unified entry model

DotCraft connects different entry points to the same project-scoped workspace, while the Unified Session Core handles execution, state, and orchestration.

Configuration

First-time setup only needs a few fields:

FieldPurposeRecommended location
ApiKeyModel API keyGlobal config
ModelDefault model nameGlobal or workspace config
EndPointOpenAI-compatible API URLGlobal or workspace config
LanguageUI language: Chinese / EnglishGlobal config
DashBoard.EnabledEnable web debugging and visual configurationWorkspace config

If unsure, put the API key globally and everything else in the workspace.

Choose the Next Step by Goal

GoalNext step
Work visually with sessions and diffsDesktop Guide
Use a full terminal interfaceTUI Guide
Share a workspace across remote or multiple clientsAppServer Mode Guide
Expose an OpenAI-compatible HTTP APIAPI Mode Guide
Connect an IDE or editorACP Mode Guide
Build bots or external adaptersSDK Overview
Run local automation tasksAutomations Guide
Inspect traces, tool calls, and merged configurationDashboard Guide

Explore More

Social Channels

DotCraft integrates with Telegram, WeChat, Feishu/Lark, QQ, WeCom, and other social channels through SDK extensions. Start with the Python SDK and TypeScript SDK.

Telegram (Python SDK)WeChat (TypeScript SDK)
Telegram channel exampleWeChat channel example

Automations

Automations are for running local workspace tasks. Scheduling, thread binding, templates, and retry flows are covered in the Automations Guide.

Desktop Automations
Desktop automations panel

Dashboard

Dashboard is DotCraft's visual inspection and configuration surface for sessions, traces, and workspace settings. See the Dashboard Guide for the full page overview.

Usage overviewSession trace
Dashboard usage overviewDashboard session trace

Advanced Topics

Troubleshooting

Desktop cannot find dotcraft

Make sure the DotCraft CLI is on PATH, or set the AppServer / dotcraft binary path in Desktop settings. Source-build users can run build.bat from the repository root first.

Model requests fail

Check that ApiKey, Model, and EndPoint belong to the same provider. OpenAI-compatible endpoints usually end with /v1.

Workspace configuration does not apply

Confirm the config is in the current workspace's .craft/config.json, then restart Desktop or the relevant host. Some AppServer and entry-point settings are read only at startup.

Apache License 2.0