⚛️ XIA — Autonomous Development Infrastructure

XIA is a powerful, locally-hosted agentic AI orchestration engine. It transforms natural language intents into directed acyclic graphs (DAGs) of tasks, dispatches them to domain-specific agent workers, and executes them with strict constitutional constraints.

🚀 Quick Start

XIA is distributed as a single, compiled cross-platform binary. You do not need Bun or Node.js installed to run it.

npm install -g @astrake/xia

1. Prerequisites

Before initializing XIA, you must have the following running locally:

• Redis (or Memurai on Windows) — Used for the BullMQ task queue.
• Qdrant — Vector database for semantic long-term memory.

You will also need API keys for your chosen AI providers (configured during xia init):

• Antigravity API key — For Gemini 2.5 Flash access.
• Kilo API key — For fast code generation (optional).

2. Initialization

Run the interactive setup wizard to configure your directories and API keys:

xia init

This will create your config directories (e.g., %APPDATA%\xia on Windows, ~/.config/xia on Linux).

3. Health Check

Verify your environment is healthy:

xia doctor

4. Start the Daemon

Start the XIA background orchestrator:

xia daemon

(You can also use xia service install to install it as a systemd service or Windows Scheduled Task).

💻 Usage

Submit tasks via natural language. The orchestrator will automatically classify the domain, plan the DAG, and dispatch the correct provider agents (Coder, Designer, Engineer).

xia run "Scaffold a new authentication module using JWT"

View the status of your tasks:

xia tasks
xia status <task_id>

Approve or reject tasks blocked by the human-approve gate:

xia approve <task_id>
xia reject <task_id> "Make the password validation stricter"

🧠 Architecture Overview

• Planner: Converts intent into a validated DAG (Directed Acyclic Graph) of tasks.
• Scheduler (BullMQ): Resolves dependencies and executes tasks in parallel.
• Context Injector: Queries Qdrant memory and injects domain-specific rules.
• Executor: Dispatches to Provider Agents and enforces gates (e.g., deterministic compile gates, HIPAA compliance gates).
• Provider Agents: Two-tier architecture (Provider → Worker). Providers orchestrate granular sub-tasks executed deterministically by workers using Bun.spawn.

📡 Integrations

• Telegram Bot: Dispatch tasks via text or voice (Whisper API) directly from your phone. Receive live worker.progress updates and inline approval buttons.
• Web Dashboard: (Coming Soon) VueFlow-based DAG visualizer and Monaco side-by-side diff reviewer.
• Push Alerts: Integration with NTFY and Pushover for FATAL daemon alerts.
• Local LLMs: Seamless failover to Ollama for sensitive domain rules (e.g., HIPAA compliance).

🔐 Domain Scoping

XIA natively supports working across different projects. Define your projects in xia.json:

{
  "projects": {
    "auruvi": {
      "path": "/home/user/projects/auruvi",
      "constitution": ["HIPAA Compliance Required", "No PII in logs"]
    },
    "hermes": {
      "path": "/home/user/projects/hermes",
      "constitution": ["HFT Latency Constraints", "Zero allocations"]
    }
  }
}

When an intent is classified into a domain, XIA automatically scopes the execution to the correct workingDir and injects the project's constitution into every agent prompt.

🤝 Contributing

See CONTRIBUTING.md for local development instructions.

📖 Detailed Setup Guide

For step-by-step installation instructions for Windows, Linux, and macOS — including Redis, Qdrant, Telegram, and background service setup — see docs/SETUP_GUIDE.md.