avp-py

Python implementation of Agent Vault Protocol
Standard conformance · Async support · Type hints

Overview

avp-py is the official Python implementation of the Agent Vault Protocol (AVP). It provides a simple, Pythonic interface for secure credential management in AI agent systems.

Features

• Standard AVP Conformance — All 7 core operations
• Multiple Backends — File, Keychain, Remote (Hardware via USB bridge)
• Async Support — Both sync and async APIs
• Type Hints — Full typing for IDE support
• Framework Integrations — LangChain, CrewAI, AutoGen ready

Installation

pip install avp-sdk

With optional backends:

pip install avp-sdk[keychain]     # OS keychain support
pip install avp-sdk[remote]       # Remote vault support (requests)
pip install avp-sdk[all]          # All optional dependencies

Quick Start

import avp

# Create vault instance
vault = avp.Vault("avp.toml")

# Authenticate
vault.authenticate()

# Store a secret
vault.store("anthropic_api_key", "sk-ant-...")

# Retrieve a secret
api_key = vault.retrieve("anthropic_api_key")

# Use with context manager (auto-cleanup)
with avp.Vault("avp.toml") as vault:
    api_key = vault.retrieve("anthropic_api_key")

Async API

import asyncio
import avp

async def main():
    async with avp.AsyncVault("avp.toml") as vault:
        await vault.authenticate()
        api_key = await vault.retrieve("anthropic_api_key")
        print(f"Retrieved key: {api_key[:10]}...")

asyncio.run(main())

Backend Selection

import avp

# File backend (encrypted)
vault = avp.Vault(backend=avp.FileBackend(
    path="~/.avp/secrets.enc",
    cipher="chacha20-poly1305"
))

# OS Keychain
vault = avp.Vault(backend=avp.KeychainBackend())

# Remote vault
vault = avp.Vault(backend=avp.RemoteBackend(
    url="https://vault.company.com",
    token="hvs.xxx"
))

Migration

import avp

# Migrate from file to keychain
avp.migrate(
    source=avp.FileBackend(path="~/.avp/secrets.enc"),
    target=avp.KeychainBackend()
)

Framework Integration

LangChain

from langchain_avp import AVPSecretManager

# Use AVP as LangChain's secret manager
secret_manager = AVPSecretManager("avp.toml")
llm = ChatAnthropic(api_key=secret_manager.get("anthropic_api_key"))

CrewAI

from crewai_avp import AVPCredentialStore

# Use AVP as CrewAI's credential store
credentials = AVPCredentialStore("avp.toml")
agent = Agent(credentials=credentials)

Environment Variable Replacement

Replace insecure .env files:

import avp

# Before: insecure .env file
# ANTHROPIC_API_KEY=sk-ant-...

# After: secure AVP vault
vault = avp.Vault("avp.toml")
os.environ["ANTHROPIC_API_KEY"] = vault.retrieve("anthropic_api_key")

Or use the AVP environment loader:

import avp

# Load all secrets into environment
avp.load_env("avp.toml", [
    "anthropic_api_key",
    "openai_api_key",
    "github_token"
])

API Reference

Vault

Conformance

Contributing

See CONTRIBUTING.md for development setup.

License

Apache 2.0 — see LICENSE.

Specification · Documentation · Issues