Project Aether 🌌

The Knowledge Computing Platform for Agentic AI.

Aether is a next-generation compiler that treats human knowledge, constraints, and architecture as source code. It parses your project's markdown and Knowledge Definition Language (.kdl) files, compiles them into a strict Enterprise Knowledge Graph (EKG), and dynamically generates hyper-optimized contexts for AI Agents at runtime.

No more RAG hallucinations. No more LLM context window bloat. Just pure, deterministic knowledge injection.

🚀 Features

• Multi-Format Compiler: Parses standard Markdown and strict .kdl files into a unified AST.
• Semantic Knowledge Graph: Discovers relationships (DependsOn, ConflictsWith) and maps them strictly using graphology.
• Optimization Engine: Automatically deduplicates rules, resolves priority conflicts (Mandatory > Recommended), and compresses context to fit LLM budgets.
• SQLite Persistence: Fully persists the compiled Knowledge Graph to a local database (.aether/graph.db).
• REST API Server: Exposes compiler and runtime via Fastify (aether serve).
• Universal Adapters: Out-of-the-box payload generation for OpenAI, Anthropic Claude, and the Model Context Protocol (MCP).

🏗️ Architecture

graph TD
    A[Markdown / KDL Docs] -->|aether compile| B(Knowledge Compiler Pipeline)
    B --> C[P1: Parser]
    C --> D[P2: Extractor]
    D --> E[P3: Relationship Discoverer]
    E --> F[P4: Ontology Mapper]
    F --> G[P7: Graph Constructor]
    G --> H[(SQLite Graph DB)]
    
    H -->|aether run| I(Aether Runtime)
    I --> J[O1: Dead Knowledge Elimination]
    J --> K[O2: Semantic Deduplication]
    K --> L[O3: Conflict Resolution]
    L --> M[O4: Context Compression]
    M --> N[CEC Assembler]
    
    N --> O{Model Adapters}
    O --> P[OpenAI]
    O --> Q[Anthropic]
    O --> R[MCP]

Loading

📦 Installation

# Clone the repository
git clone https://github.com/hacktronaut/project-aether.git
cd project-aether

# Install dependencies (requires Node.js 20+)
npm install

# Build the monorepo
npm run build

🛠️ Quick Start

1. Compile your Knowledge Base

Point Aether at a directory containing your project's markdown or .kdl specifications:

npx aether compile ./docs -o ./.aether/graph.db

2. Run a Mission

Execute a natural language mission against your compiled knowledge graph. Aether will traverse the graph, optimize the context, and output a Compiled Execution Context (CEC) payload for your LLM:

npx aether run --mission "Implement the new JWT authentication flow" --model gpt-4o --budget 20

3. Start the REST API (Optional)

Run Aether as a standalone service for your AI agents to query dynamically:

npx aether serve -p 3000

• POST /compile: Compile a new set of documents.
• POST /mission: Run a mission and return the optimized CEC.

💻 CLI Reference

aether compile <source-dir>

Compile knowledge sources from a directory into a serialized Knowledge Graph.

Arguments:

• <source-dir>: Directory containing markdown/KDL knowledge sources.

Options:

• -o, --output <path>: Path to write the compiled graph database (default: ./.aether/graph.db).
• --strict: Fail compilation on any warning or error (default: false).

aether run

Run a mission against the compiled Knowledge Graph to produce a CEC and model payload.

Options:

• --mission <text>: (Required) The natural language mission or SDD task objective.
• -g, --graph <path>: Path to the compiled graph database (default: ./.aether/graph.db).
• -m, --model <name>: Target model name (default: gpt-4o).
• -b, --budget <count>: Context compression budget, i.e., max non-mandatory nodes (default: 10).

aether serve

Start the Aether REST API server.

Options:

• -p, --port <number>: Port to listen on (default: 3000).
• -g, --graph <path>: Path to the compiled graph database (default: ./.aether/graph.db).

📁 Monorepo Structure

• @aether/core: The heart of the system (Compiler, Runtime, Optimizations, Graph).
• @aether/cli: The command-line interface and Fastify REST API.
• @aether/adapters: Payload renderers for OpenAI, Anthropic, and MCP.
• @aether/kdl: Parser for the Knowledge Definition Language.

🤝 Contributing

Contributions are welcome! Please see CONTRIBUTING.md for guidelines. Run npm test to ensure all integration and unit tests pass before submitting a PR.