English · 简体中文 · 繁體中文 · 日本語 · 한국어 · Deutsch · Français · Español · Português (Brasil) · Русский
HagiCode keeps AI from losing context while it changes your code: understand the repository first, set the scope, then execute with a trail you can review.
An AI workbench for project development. It plugs into the Agent CLIs and models you already use, so you don't have to change editors.
Install · Product Overview · Website · Microsoft Store · Blog
- Mainstream Agent CLIs in one workbench, such as Claude Code, Codex, GitHub Copilot, OpenCode, and more. See the supported list.
- Windows, macOS, and Linux, as a desktop app or a self-hosted container.
- Free core features. Desktop, Container, and Microsoft Store are three ways to get HagiCode, not three paid tiers.
- Read-only by default. A normal session only reads your code; switch to edit mode when you want changes.
1. Pick where it runs.
| You work on | Choose | Needs on the host | |
|---|---|---|---|
| Windows | Microsoft Store | Node.js 24+ and Git | Get it |
| macOS, Linux, or Windows without the Store | Desktop | Node.js 24+ and Git | Download |
| A server, NAS, or a team | Container (Docker Compose) | Docker. No Node.js or Git needed. | Deploy |
Step-by-step instructions are in the installation guide.
2. Set up. Finish the first-run wizard and create at least one project.
3. Start a session. Open a read-only normal session to understand the project. When you're ready to change code, switch to edit mode or start a proposal session.
- Understand first. Analyze existing code in a read-only normal session.
- Set the scope. For complex changes, pick the project and repositories in a proposal session, then agree on goals, tasks, and validation before anything is edited.
- Execute. Switch to edit mode and work against the goals, tasks, and validation you agreed.
- Review and keep the context. Check sessions and execution state in the workbench, and let AI Compose Commit organize the commits. Proposals and commit records stay available for later, and MonoSpecs coordinates changes across repositories.
You don't need a proposal every time. For a quick question about the code, start with a normal session.
- Proposal-driven AI coding:
OpenSpeckeeps scope, tasks, impact analysis, validation, and execution connected instead of turning complex work into blind edits. - Bring your own Agent CLI and model:
OmniRoutekeeps the CLI choice separate from the model and subscription layer, so you can route models and endpoints without being tied to one default stack. - A full development workspace:
MonoSpecsfor multi-repository scope,Skillsfor installable extensions,Vaultfor reusable knowledge,AI Compose Commitandcode-serverfor finishing the job in the same place. - Gamified feedback with real signals: achievements, daily reports, efficiency multipliers, and token throughput make long-running AI work easier to read. It is part of the workbench, and you can ignore it.
Tools HagiCode builds on or connects to. Most are optional and can be turned on when you need them. The Settings → Integration tab holds Git, GitHub, OmniRoute, TokenMonitor, and Code Server.
| Area | Component | What it does in HagiCode |
|---|---|---|
| Agents and models | Agent CLIs | Claude Code, Codex, GitHub Copilot, OpenCode, and more run inside the workbench. |
| OmniRoute | Shared model gateway. Set an endpoint and choose which Agent CLIs use its model catalog. | |
| Specs and scope | OpenSpec | Proposal-driven workflow that connects scope, tasks, validation, and execution. |
| MonoSpecs | Manages repository relationships and change scope across several repositories. | |
| Code and version control | Git | Git Profiles give AI commits an explicit author. Optional status monitoring for active projects. |
| GitHub | Reuses your local gh session to sync sessions and import GitHub data. The token stays in memory only. |
|
| code-server | Browser editing entry, so you can finish a change without leaving the workbench. Settings shows its configuration and status, read-only. | |
| Knowledge and extensions | Vault | Collects reference repositories, folders, and notes for reuse in later work. |
| Skills & HagiTask | Browse the skills gallery and install skills, or install HagiTask presets from the community gallery. | |
| Input and feedback | TokenMonitor | Shows today's token usage and cost from the TokenMonitor desktop app in the status bar. Add a HostHub address and secret for near-real-time updates. |
| Doubao voice input | Dictate proposals, comments, and session messages with Doubao (豆包) speech recognition. |
- A good fit: developers and teams who maintain several repositories and want complex changes planned before they run, or who already use Claude Code, Codex, or another Agent CLI and want them in one workbench.
- Not required: if you only want a few lines of completion in your editor, your current tools are enough.
Plus and DLC details live on the docs site. The Steam main application has been retired; see the Steam status FAQ.
- Hagicode Plus — bundle and upgrade guidance
- Turbo Engine DLC — higher proposal concurrency, plus document themes, custom signature, branding, and avatars (Microsoft Store, one-time purchase)
- site - Official marketing site and public GitHub landing repository
- desktop - Desktop client for local-first HagiCode workflows
- container - Self-hosted deployment path for teams and server environments
- docs - Documentation site with the canonical product overview and guides
| Package | What it does | npm |
|---|---|---|
@hagicode/cli |
Manage projects, proposals, vaults, chat, and AutoTask from the terminal. |  |
@hagicode/hagiscript |
Hagiscript language tooling and runtime management. |  |
@hagicode/hagi18n |
Audit, sync, and manage locale resources. |  |
@hagicode/imgbin |
Generate, annotate, and index image assets. |  |
@hagicode/skillsbase |
Bootstrap and maintain managed agent skills repositories. |  |
Ready to explore the full product story? Start at hagicode.com and choose the path that fits your workflow.