Files
fusion/README.md

143 lines
5.4 KiB
Markdown

# hai
AI-orchestrated task board. Like Trello, but your tasks get specified, executed, and delivered by AI — powered by [pi](https://github.com/badlogic/pi-mono).
## Workflow
```
┌──────────┐ ┌──────────┐ ┌────────────┐ ┌───────────┐ ┌──────┐
│ Triage │───▶│ Todo │───▶│ In Progress│───▶│ In Review │───▶│ Done │
│ │ │ │ │ │ │ │ │ │
│ raw idea │ │ AI spec'd│ │ AI working │ │ ready to │ │merged│
│ │ │ & ready │ │ in worktree│ │ merge │ │ │
└──────────┘ └──────────┘ └────────────┘ └───────────┘ └──────┘
pi ▲ pi human
specifies deps gate executes reviews
```
1. **Triage** — Throw rough ideas in. Pi picks them up and writes a proper task spec.
2. **Todo** — Fully specified, ready to go. Scheduler moves them when deps are met.
3. **In Progress** — Pi works the task in an isolated git worktree.
4. **In Review** — Work is done. Merge the worktree and close. Toggle **Auto-merge** in the column header to automatically merge tasks as they arrive.
5. **Done** — Shipped.
Tasks with dependencies are processed sequentially. Independent tasks run in parallel.
## Quick Start
```bash
# Install dependencies
pnpm install
# Start the board (UI only)
pnpm dev dashboard
# Start with AI engine (auto-specify + auto-execute)
pnpm dev dashboard -- --engine
# Create a task via CLI
pnpm dev task create "Fix the login redirect bug"
# List tasks
pnpm dev task list
# Move a task
pnpm dev task move HAI-001 todo
# Attach a screenshot to a task
pnpm dev task attach HAI-001 ./screenshot.png
```
Then open [http://localhost:4040](http://localhost:4040).
## Prerequisites
The AI engine uses [pi](https://github.com/badlogic/pi-mono) agent sessions under the hood. You need:
1. **pi installed:** `npm install -g @mariozechner/pi-coding-agent`
2. **API key configured:** Run `pi` and use `/login` or set `ANTHROPIC_API_KEY`
hai reuses your existing pi authentication — no separate setup needed.
## Packages
| Package | Description |
|---------|-------------|
| `@hai/core` | Domain model — tasks, board columns, file-based store |
| `@hai/dashboard` | Web UI — Express server + kanban board with SSE |
| `@hai/engine` | AI engine — triage (pi), execution (pi + worktrees), scheduling |
| `hai` (cli) | CLI — `hai dashboard`, `hai task create/list/move/attach` |
## Architecture
### Task Storage
Tasks live on disk in `.hai/tasks/` in the project root:
```
.hai/
├── config.json # Board config + ID counter
└── tasks/
└── HAI-001/
├── task.json # Metadata (column, deps, timestamps)
├── PROMPT.md # Task specification
└── attachments/ # Screenshot images (optional)
```
### Board UI
Real-time kanban board at `localhost:4040`:
- Drag-and-drop cards between columns
- Create tasks from the web UI
- Click cards for detail view with move/delete actions
- Server-Sent Events for live updates across tabs
### API Rate Limiting
All API endpoints (`/api/*`) are rate limited to prevent abuse. Limits are applied per client IP:
| Scope | Limit | Window |
|-------|-------|--------|
| General API (`/api/*`) | 100 requests | 1 minute |
| SSE connections (`/api/events`) | 10 connections | 1 minute |
Every API response includes standard rate limit headers:
| Header | Description |
|--------|-------------|
| `RateLimit-Limit` | Maximum requests allowed per window |
| `RateLimit-Remaining` | Requests remaining in the current window |
| `RateLimit-Reset` | Seconds until the rate limit window resets |
| `Retry-After` | Seconds to wait before retrying (only on 429 responses) |
When a client exceeds the limit, the API returns `429 Too Many Requests`.
### AI Engine (`--engine`)
When enabled, three components run:
- **TriageProcessor** — Watches triage column. Spawns a pi agent session that reads the project, understands context, and writes a full PROMPT.md specification. Moves task to todo.
- **Scheduler** — Watches todo column. Resolves dependency graphs. Moves tasks to in-progress when deps are satisfied and concurrency allows (default: 2 concurrent). When `groupOverlappingFiles` is enabled in settings, tasks whose `## File Scope` sections share files are serialized to prevent merge conflicts.
- **TaskExecutor** — Listens for tasks entering in-progress. Creates a git worktree, spawns a pi agent session with full coding tools scoped to the worktree, and executes the specification. Moves to in-review on completion.
Each pi agent session gets:
- Custom system prompt for its role (triage specifier vs task executor)
- Tools scoped to the correct directory (`createCodingTools(cwd)`)
- In-memory sessions (no persistence needed)
- The user's existing pi auth (API keys from `~/.pi/agent/auth.json`)
## Development
```bash
pnpm install
pnpm dev dashboard # Board only
pnpm dev dashboard -- --engine # Board + AI engine
pnpm dev task list # CLI commands
```
## License
ISC