ADP Hypervisor

A server implementation of the ADP (Agentic Data Protocol) that enables AI Agents to safely access heterogeneous data systems through a unified Intent-based interface.

Features

• Intent-based Data Access: Express data operations as high-level intents (LOOKUP, QUERY, INGEST, REVISE) instead of raw queries
• Heterogeneous Data Sources: Unified interface to access relational databases, vector stores, and more
• Contract-driven Interaction: Discover available resources, describe usage contracts, validate before execute
• Extensible Architecture: Pluggable backends and manifest providers for easy customization

Requirements

• Python 3.11+
• uv (recommended package manager)

Installation

Using uv (Recommended)

# Install uv if not already installed
curl -LsSf https://astral.sh/uv/install.sh | sh

# Clone the repository
git clone https://github.com/agenticdataprotocol/adp-hypervisor.git
cd adp-hypervisor

# Install dependencies
uv sync

# Install with development dependencies
uv sync --extra dev

Using pip

pip install -e ".[dev]"

Quick Start

1. Prepare Manifest Files

Create a configuration directory with three YAML manifest files:

mkdir -p my-config
cp conf/physical.yaml.template my-config/physical.yaml
cp conf/semantic.yaml.template my-config/semantic.yaml
cp conf/policy.yaml.template my-config/policy.yaml

Edit each file to configure your backends, resources, and policies. See the templates in conf/ for detailed examples, or jump straight to the examples directory for a ready-to-run demo.

2. Start the Server

python -m adp_hypervisor --config my-config

The server starts in stdio mode, reading JSON-RPC requests from stdin and writing responses to stdout.

CLI Usage

python -m adp_hypervisor --config <path> [--log-level LEVEL] [--transport TYPE]

Option	Default	Description
--config	Required	Path to manifest directory (physical.yaml, semantic.yaml, policy.yaml)
--log-level	From logging config	Override root log level: DEBUG, INFO, WARNING, ERROR, CRITICAL
--transport	stdio	Transport type: stdio (HTTP planned for future release)

Logging Configuration

Logging is configured via logging_conf.yaml in the config directory. The file follows Python's standard logging.config.dictConfig format.

Copy the provided template to get started:

cp conf/logging_conf.yaml.template <config-dir>/logging_conf.yaml

Default behaviour (used when no logging_conf.yaml is present):

• Logs go to both stderr (console) and a rotating file at ./hypervisor-logs/hypervisor.log
• The log directory is created automatically if it does not exist
• The resolved log file path is printed to stderr at startup
• Root log level: INFO
• File rotation: 10 MB per file, 5 backup files

Key tunable fields in logging_conf.yaml:

Field	Default	Description
root.level	INFO	Root log level (overridable with --log-level)
handlers.file_handler.filename	./hypervisor-logs/hypervisor.log	Log file path (relative to CWD)
handlers.file_handler.maxBytes	10485760 (10 MB)	Max file size before rotation
handlers.file_handler.backupCount	5	Number of backup files to keep

Note: Logs must never go to stdout when using stdio transport — that stream is reserved for JSON-RPC responses.

3. Send a Request

With the server running, send a JSON-RPC request via stdin:

echo '{"jsonrpc":"2.0","id":1,"method":"adp.initialize","params":{"protocolVersion":"2026-01-20","capabilities":{},"clientInfo":{"name":"test","version":"1.0"}}}' \
  | python -m adp_hypervisor --config my-config

Programmatic Usage

import asyncio
from adp_hypervisor import ADPServer
from adp_hypervisor.manifest.yaml_provider import YamlManifestProvider
from pathlib import Path

provider = YamlManifestProvider(
    physical_path=Path("my-config/physical.yaml"),
    semantic_path=Path("my-config/semantic.yaml"),
    policy_path=Path("my-config/policy.yaml"),
)
server = ADPServer(manifest_provider=provider)
asyncio.run(server.run())

Examples

The examples directory contains a ready-to-run demo with Docker Compose infrastructure and pre-configured manifests. See examples/README.md for the full walkthrough.

Backend	Status	Description
PostgreSQL	✅ Implemented	E-commerce dataset (customers, products, orders) with LOOKUP and QUERY intents
pgvector	✅ Implemented	Vector similarity search demo with embedded product catalog items
MongoDB	✅ Implemented	User profile collection with LOOKUP and QUERY intents in the shared examples demo
Local Filesystem	✅ Implemented	Invoice files organised by fulfilment status; supports LOOKUP, QUERY, INGEST, REVISE (no Docker required)

Development

Running Tests

# Unit tests
uv run python -m unittest discover -s tests/unit -t tests -v

# Integration tests (requires Docker for testcontainers)
uv run python -m unittest discover -s tests/integration -v

# All tests
uv run python -m unittest discover -s tests -t tests -v

Code Formatting

uv run black .

Linting

uv run ruff check .

Type Checking

uv run mypy src/

Project Structure

adp-hypervisor/
├── pyproject.toml              # Project configuration
├── conf/                       # Manifest templates (physical, semantic, policy, logging)
├── examples/                   # Ready-to-run demo (Docker + manifests)
├── src/adp_hypervisor/         # Main hypervisor package
│   ├── server.py               # ADPServer main class
│   ├── __main__.py             # CLI entry point
│   ├── transport/              # Transport layer (stdio; HTTP planned)
│   ├── protocol/               # JSON-RPC types, errors, dispatcher
│   ├── handlers/               # ADP method handlers
│   ├── manifest/               # Manifest models and providers
│   └── policy/                 # ACCESS policy enforcement and RBAC
├── src/backends/               # Backend implementations (RDBMS, etc.)
└── tests/                      # Test suite
    ├── unit/                   # Unit tests
    └── integration/            # Integration & E2E tests

License

Apache-2.0