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:
| Platform | Recommended file |
|---|---|
| Windows | DotCraft-Desktop-win-x64-Setup.exe |
| macOS | DotCraft-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.

If you prefer building from source, install the .NET 10 SDK, Rust toolchain, and Node.js, then run this from the repository root:
build.batOn Linux / macOS, run:
bash build_linux.bat2. 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/:
dotcraftStart from a real project folder instead of an empty directory so the agent can read repository structure, existing docs, and build scripts.

3. Configure a Model
DotCraft supports two common model paths:
| Path | Best for |
|---|---|
| OpenAI-compatible API key | OpenAI API, OpenRouter, DeepSeek, and compatible providers |
| CLIProxyAPI | Reusing a local coding-agent CLI through an OpenAI-compatible proxy |

The minimal configuration usually looks like this:
{
"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:
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:
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.
| Dimension | Gateway | Unified Session Core |
|---|---|---|
| Client customization | Hard to customize once everything is flattened into a message bus | Flexible, native client experiences |
| Approval / HITL | Cannot express platform-native approval flows | Rendered with native platform UI |
| Cross-channel resume | Not supported | Conversations can resume across channels |
| Workspace persistence | Not supported | Designed around the workspace |

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:
| Field | Purpose | Recommended location |
|---|---|---|
ApiKey | Model API key | Global config |
Model | Default model name | Global or workspace config |
EndPoint | OpenAI-compatible API URL | Global or workspace config |
Language | UI language: Chinese / English | Global config |
DashBoard.Enabled | Enable web debugging and visual configuration | Workspace config |
If unsure, put the API key globally and everything else in the workspace.
Choose the Next Step by Goal
| Goal | Next step |
|---|---|
| Work visually with sessions and diffs | Desktop Guide |
| Use a full terminal interface | TUI Guide |
| Share a workspace across remote or multiple clients | AppServer Mode Guide |
| Expose an OpenAI-compatible HTTP API | API Mode Guide |
| Connect an IDE or editor | ACP Mode Guide |
| Build bots or external adapters | SDK Overview |
| Run local automation tasks | Automations Guide |
| Inspect traces, tool calls, and merged configuration | Dashboard 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) |
|---|---|
![]() | ![]() |
Automations
Automations are for running local workspace tasks. Scheduling, thread binding, templates, and retry flows are covered in the Automations Guide.
| Desktop Automations |
|---|
![]() |
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 overview | Session trace |
|---|---|
![]() | ![]() |
Advanced Topics
- Use Hooks to run scripts on lifecycle events.
- Use sandbox and security settings to constrain file, shell, and network access.
- Use the Workspace Sample to validate a complete workspace template.
- Return to the Documentation Index and choose by goal.
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.




