An AI coding agent for your desktop and terminal.
Understand a codebase, make changes, run commands, and keep development work organized in persistent tasks.
English | 中文
Interfaces • Quick start • Capabilities • Documentation
ACECode is a repository-aware AI coding agent with two first-class interfaces: a visual Desktop app and a keyboard-first terminal TUI. Both use the same agent core, model profiles, permission system, and session storage.
Choose a workspace, start or revisit tasks, select a model and permission mode, attach context, and follow the agent's work from one window.
Work directly from a project terminal with streaming responses, tool progress, session controls, and a focused keyboard-driven interface.
- Open ACECode Desktop.
- On first use, open Settings > Models and configure a provider and model.
- Select New task, then choose an existing workspace, open a local project, or continue with No workspace for a general task.
- Choose the model and permission mode below the prompt.
- Describe the result you want and press
Enter.
Tip
Type @ to reference a workspace file or directory. In Desktop, the add button also passes local files, including images and large files, to the agent as on-demand path references.
Configure a model the first time you use ACECode:
acecode configure
Start ACECode from the repository you want it to work in:
cd /path/to/your/project
acecodeType a concrete request and press Enter:
Explore this repository, explain how sessions are stored, then add a focused test for the serializer.
Resume the most recent task for the current project when you return:
acecode --resume
Important
In the default permission mode, ACECode normally reads project context automatically and asks before sensitive writes or command execution. Review permission requests, tool output, and file changes carefully.
Explain the architecture and point me to the main entry points.Find the cause of this failing test and propose a minimal fix.Refactor @src/domain/session/ without changing its public behavior, then run the focused tests.Review my current diff for correctness, regressions, and missing tests.
The best requests name the desired outcome, relevant files or constraints, and how the result should be verified.
- Understand repositories — search code, follow call paths, inspect configuration, and explain unfamiliar systems.
- Change code safely — edit files, show diffs, run commands and tests, and ask for permission when required.
- Keep context — persist tasks and sessions, resume earlier work, compact long conversations, and revisit history.
- Choose your interface — manage multiple workspaces visually in Desktop or stay close to the shell in the TUI.
- Use your models — work with GitHub Copilot, OpenAI-compatible APIs, Anthropic, and saved model profiles.
- Extend the agent — add reusable Skills, MCP servers, tools, hooks, and connectors for specialized workflows.
| Command | Purpose |
|---|---|
/help |
Show the commands available in the installed version. |
/model |
Inspect or switch the current model. |
/resume |
Open the session picker. |
/skills |
Open the Skills capability center. |
/mcp |
Open the MCP server capability center. |
/exit |
Leave ACECode. |
- User manual — everyday TUI workflows, permissions, sessions, Skills, and MCP.
- Architecture — runtime surfaces and source ownership.
- Skills guide — create and use reusable workflows.
- Desktop workspaces — workspace and task behavior in the desktop app.
- Linux self-update — updater ZIPs, one-time legacy bootstrap, and release verification.
- UOS / Deepin builds — optional DTK frame effects, WebKitGTK 4.0, and dedicated release packages.
This project is licensed under the MIT License.