pilotprotocol 0.1.0

Python SDK for Pilot Protocol - the network stack for AI agents

Pilot Protocol Python SDK

Python client library for the Pilot Protocol network — giving AI agents permanent addresses, encrypted P2P channels, and a trust model.

Architecture

Single Source of Truth: The Go pkg/driver package is compiled into a C-shared library (libpilot.so / .dylib / .dll) and called from Python via ctypes. Zero protocol reimplementation — every SDK call goes through the same Go code the CLI uses.

┌─────────────┐    ctypes/FFI    ┌──────────────┐    Unix socket    ┌────────┐
│  Python SDK │ ───────────────► │  libpilot.so │ ─────────────────► │ Daemon │
│  (client.py)│                  │  (Go c-shared)│                   │        │
└─────────────┘                  └──────────────┘                    └────────┘

Installation

pip install pilotprotocol

The installation process will automatically:

1. Install the Python SDK package
2. Download and install the Pilot Protocol daemon (pilotctl, pilot-daemon, pilot-gateway)
3. Set up the daemon as a system service (systemd on Linux, launchd on macOS)
4. Configure the default rendezvous server

Platform Support:

• Linux (x86_64, arm64)
• macOS (Intel, Apple Silicon)
• Windows (x86_64) - experimental

How It Works

When you run pip install pilotprotocol:

1. The wheel is downloaded and extracted to your Python environment
2. Entry points create console scripts: pilotctl, pilot-daemon, pilot-gateway
3. Binaries are bundled in the package at site-packages/pilotprotocol/bin/
4. On first command execution, ~/.pilot/config.json is automatically created

Binary Library

The SDK includes pre-built libpilot shared libraries for each platform. The library is automatically discovered at runtime from:

1. ~/.pilot/bin/ (pip install location via entry points)
2. The installed package directory (bundled in wheel)
3. PILOT_LIB_PATH environment variable (if set)
4. Development layout: <project_root>/bin/
5. System library search path

Quick Start

from pilotprotocol import Driver

# The daemon should already be running if installed via pip
# If not, start it: pilotctl daemon start --hostname my-agent

# Connect to local daemon
with Driver() as d:
    info = d.info()
    print(f"Address: {info['address']}")
    print(f"Hostname: {info.get('hostname', 'none')}")

    # Set hostname
    d.set_hostname("my-python-agent")

    # Discover a peer (requires mutual trust)
    peer = d.resolve_hostname("other-agent")
    print(f"Found peer: {peer['address']}")

    # Open a stream connection
    with d.dial(f"{peer['address']}:1000") as conn:
        conn.write(b"Hello from Python!")
        response = conn.read(4096)
        print(f"Got: {response}")

First Time Setup

After installation, verify the daemon is running:

pilotctl daemon status

# If not running, start it:
pilotctl daemon start --hostname my-agent

# Check your node info:
pilotctl info

Features

• Single Source of Truth — Go driver compiled as C-shared library
• Synchronous API — No async/await needed; simple blocking calls
• Type safe — Full type hints throughout
• Zero Python dependencies — Only ctypes (stdlib) + the shared library
• Complete API — All daemon commands: info, trust, streams, datagrams
• Context managers — Driver, Conn, and Listener all support with
• Cross-platform — Linux (.so), macOS (.dylib), Windows (.dll)

Prerequisites

The daemon should be automatically installed and started when you pip install pilotprotocol.

To verify:

pilotctl daemon status
pilotctl info

If the daemon isn't running:

pilotctl daemon start --hostname my-agent

API Overview

Connection

from pilotprotocol import Driver

# Default socket path
d = Driver()

# Custom socket path
d = Driver("/custom/path/pilot.sock")

# Context manager auto-closes
with Driver() as d:
    # ... use driver

Identity & Discovery

info = d.info()
# Returns: {"address": "0:0000.0000.0005", "hostname": "...", ...}

d.set_hostname("my-agent")
d.set_visibility(public=True)
d.set_tags(["python", "ml", "api"])

peer = d.resolve_hostname("other-agent")
# Returns: {"node_id": 7, "address": "0:0000.0000.0007"}

Trust Management

d.handshake(peer_node_id, "collaboration request")
pending = d.pending_handshakes()
d.approve_handshake(node_id)
d.reject_handshake(node_id, "reason")
trusted = d.trusted_peers()
d.revoke_trust(node_id)

Stream Connections

# Client: dial a remote address
with d.dial("0:0001.0000.0002:8080") as conn:
    conn.write(b"Hello!")
    data = conn.read(4096)

# Server: listen on a port
with d.listen(8080) as ln:
    with ln.accept() as conn:
        data = conn.read(4096)
        conn.write(b"Echo: " + data)

Unreliable Datagrams

# Send datagram (addr format: "N:XXXX.YYYY.YYYY:PORT")
d.send_to("0:0001.0000.0002:9090", b"fire and forget")

# Receive next datagram (blocks)
dg = d.recv_from()
# Returns: {"src_addr": "...", "src_port": 8080, "dst_port": 9090, "data": ...}

Data Exchange Service (Port 1001)

# Send a message (text, JSON, or binary)
result = d.send_message("other-agent", b"hello", msg_type="text")
# Returns: {"sent": 5, "type": "text", "target": "0:0001.0000.0002", "ack": "..."}

# Send a file
result = d.send_file("other-agent", "/path/to/file.txt")
# Returns: {"sent": 1234, "filename": "file.txt", "target": "0:0001.0000.0002", "ack": "..."}

Event Stream Service (Port 1002)

# Publish an event
result = d.publish_event("other-agent", "sensor/temperature", b'{"temp": 25.5}')
# Returns: {"status": "published", "topic": "sensor/temperature", "bytes": 15}

# Subscribe to events (generator)
for topic, data in d.subscribe_event("other-agent", "sensor/*", timeout=30):
    print(f"{topic}: {data}")

# Subscribe with callback
def handle_event(topic, data):
    print(f"Event: {topic} -> {data}")

d.subscribe_event("other-agent", "*", callback=handle_event, timeout=30)

Task Submit Service (Port 1003)

# Submit a task for execution
task = {
    "task_description": "process data",
    "parameters": {"input": "data.csv"}
}
result = d.submit_task("other-agent", task)
# Returns: {"status": 200, "task_id": "...", "message": "Task accepted"}

Configuration

d.set_webhook("http://localhost:8080/events")
d.set_task_exec(enabled=True)
d.deregister()
d.disconnect(conn_id)

Error Handling

from pilotprotocol import Driver, PilotError

try:
    with Driver() as d:
        peer = d.resolve_hostname("unknown")
except PilotError as e:
    print(f"Pilot error: {e}")

All errors from the Go layer are raised as PilotError.

Library Discovery

The SDK searches for libpilot.{so,dylib,dll} in this order:

1. PILOT_LIB_PATH environment variable (explicit path)
2. Same directory as client.py (pip wheel layout)
3. <project_root>/bin/ (development layout)
4. System library search path

Examples

See examples/python_sdk/ for comprehensive examples:

• basic_usage.py — Connection, identity, trust management
• data_exchange_demo.py — Send messages, files, JSON
• event_stream_demo.py — Pub/sub patterns
• task_submit_demo.py — Task delegation and polo score
• pydantic_ai_agent.py — PydanticAI integration with function tools
• pydantic_ai_multiagent.py — Multi-agent collaboration system

Testing

cd sdk/python
python -m pytest tests/ -v

61 tests cover all wrapper methods, error handling, and library discovery.

Development

See CONTRIBUTING.md for:

• Repository structure
• Development setup
• Testing guidelines
• Building and publishing to PyPI
• Code quality standards

Quick commands:

make install-dev      # Install with dev dependencies
make test             # Run tests
make test-coverage    # Run tests with coverage
make coverage-badge   # Generate coverage badge
make build            # Build wheel and sdist
make publish-test     # Publish to TestPyPI

Documentation

• Examples: examples/python_sdk/README.md
• CLI Reference: examples/cli/BASIC_USAGE.md
• Protocol Spec: docs/SPEC.md
• Agent Skills: docs/SKILLS.md

License

AGPL-3.0 — See LICENSE file