abir-guard 3.1.1

PQC Agent Memory Vault - Quantum-resistant vault for AI Agent memory

Abir-Guard v3.1.0 — Quantum-Resilient Agentic Vault for AI Agent Memory

Protect AI agent secrets, API keys, and memory with NIST-standard post-quantum cryptography.

---

- Legacy memory storage is a ticking time bomb.
+ Abir-Guard: The Quantum-Resilient "Black Box" for Autonomous Agent Memory.

---

At a Glance

Category	Highlights
Quantum-Safe	ML-KEM-1024 (production-ready), ML-DSA-65, X25519 hybrid KEM, AES-256-GCM envelope encryption
Multi-Language	Python SDK, Rust CLI + Library, Go SDK, JavaScript SDK
AI Native	LangChain tools, CrewAI agents, MCP JSON-RPC server, HTTP MCP API
Hardened	FIPS 140-3 mode, key revocation (CRL), auto rotation, remote attestation, differential privacy, canary honeypots, tamper-evident audit logs
Hardware Ready	YubiKey/FIDO2, TPM 2.0 seal/unseal, Apple Secure Enclave, Intel SGX detection, HSM integration, zero-copy memory policy, Argon2id KDF (OWASP params)
Tested	109 unit tests pass across all languages, CI/CD pipeline, dependabot

---

System Architecture

graph TB
    subgraph "AI Agent"
        A1[LangChain Agent]
        A2[CrewAI Agent]
        A3[Custom MCP Client]
    end

    subgraph "Abir-Guard Vault"
        subgraph "Encryption Engine"
            E1[AES-256-GCM]
            E2[X25519 ECDH]
            E3[ML-KEM-1024]
            E4[HKDF / Argon2id]
        end

        subgraph "Security Layer"
            S1[Key Revocation CRL]
            S2[Auto Key Rotation]
            S3[FIPS 140-3 Mode]
            S4[Differential Privacy]
            S5[Remote Attestation]
            S6[Canary Honeypots]
        end

        subgraph "Persistence"
            P1[Encrypted Disk Vault]
            P2[SHAMIR Secret Sharing]
            P3[HSM / TPM Storage]
        end

        subgraph "Signatures"
            G1[ML-DSA-65 Signing]
            G2[Audit Hash Chain]
        end
    end

    A1 -->|encrypt/decrypt| E1
    A2 -->|encrypt/decrypt| E1
    A3 -->|JSON-RPC| E1

    E1 --> S1
    E1 --> S2
    E1 --> S3
    E1 --> S4
    E1 --> S5
    E1 --> S6

    S1 --> P1
    S2 --> P1
    S3 --> P1
    S4 --> P1

    P1 --> G2
    P2 --> P3
    P1 --> G1

---

Table of Contents

• Overview
• Use Cases
• Prerequisites & Installation
• Quick Start
• Python SDK Guide
• Rust CLI & Library Guide
• Go SDK Guide
• JavaScript SDK Guide
• MCP Server Guide
• LangChain & CrewAI Integration
• Docker Deployment
• HSM & TPM Integration
• Quantum Readiness
• Security Architecture
• Run Tests
• Project Structure
• Roadmap
• Contributing
• License
• Developer

---

Overview

Abir-Guard is a production-grade, quantum-resistant encryption vault built for AI agent memory and sensitive data protection. It implements NIST-standard Post-Quantum Cryptography (PQC) to defend against Harvest Now, Decrypt Later (HNDL) attacks — where adversaries collect encrypted data today to decrypt once quantum computers become powerful enough.

Three-phase implementation:

1. Bedrock — Hybrid KEM, AES-256-GCM, zero-copy memory, MCP server, LangChain/CrewAI SDKs
2. Hardware & Security — ML-DSA-65 signatures, SHAMIR secret sharing, Argon2id KDF, HSM/TPM
3. Ecosystem & Hardening — Key revocation (CRL), auto rotation, FIPS 140-3 mode, differential privacy, remote attestation, Go SDK

Written in three languages for maximum portability: Python for AI agent integration, Rust for high-performance cryptography, Go for infrastructure tooling, and JavaScript for browser and Node.js environments.

---

Use Cases

1. Protecting AI Agent API Keys

AI agents frequently handle API keys, OAuth tokens, and service credentials. Abir-Guard encrypts these at rest and in memory, ensuring no raw secrets leak through memory dumps, swap files, or process inspection.

vault.store("gpt_agent", b"OPENAI_API_KEY=sk-...")

2. Secure Agent-to-Agent Communication

Use the MCP JSON-RPC server as a local encryption gateway. Agents send plaintext to the server; encryption happens locally without exposing data to LLM context or consuming tokens.

3. Regulatory Compliance (FIPS 140-3)

Enable strict FIPS mode to enforce NIST-approved algorithms only, block non-compliant fallbacks, enforce minimum key lengths, and maintain audit trails for compliance audits.

from abir_guard.fips_mode import FIPSEncryptor
fips = FIPSEncryptor()
result = fips.encrypt(data, key)

4. Multi-Party Key Recovery (SHAMIR)

Split master secrets across trusted parties. A 3-of-5 SHAMIR scheme means any three administrators can reconstruct the master key, but fewer than three learn nothing.

./target/release/abir-guard shamir-split "master-key" -t 3 -n 5

5. Breach Detection via Canary Keys

Plant honeypot keys that alert when accessed. If an attacker compromises the vault and uses a canary key, you know immediately.

canary_id = vault.add_canary()
if vault.check_canary():
    alert_security_team("Breach detected!")

6. Automatic Key Lifecycle Management

Configure time-based or usage-based key rotation. Keys automatically expire after N operations or N hours, reducing the window of exposure from compromised keys.

rotation_manager.register_key("agent-1", max_operations=1000)

7. Remote Attestation for Decryption Gates

Before decrypting sensitive data, verify the runtime environment is untampered — checking binary integrity, environment variables, and memory canaries.

from abir_guard.attestation import AttestationVerifier
verifier = AttestationVerifier()
proof = IntegrityProof()
proof.compute(challenge)
if not verifier.verify_proof(proof.to_dict()):
    raise Exception("Runtime integrity check failed")

8. Post-Quantum Digital Signatures (ML-DSA-65)

Sign and verify data integrity using NIST FIPS 204 ML-DSA-65 — post-quantum secure, tamper-evident, non-repudiation guarantees.

let keypair = ml_dsa::generate_keypair().unwrap();
let signature = ml_dsa::sign(b"agent-memory", &keypair.signing_key).unwrap();

---

Prerequisites & Installation

System Requirements

Component	Minimum	Recommended
OS	Linux, macOS, Windows	Ubuntu 22.04+, macOS 13+, Windows 11
CPU	x86_64, ARM64	Any modern multi-core
RAM	128 MB	256 MB+ (Argon2id uses 64 MB)
Disk	50 MB	100 MB

Python SDK

# Prerequisites: Python 3.10+
python3 --version  # Must be 3.10 or higher

# Install package and dev dependencies
pip install -e ".[dev]"

# Optional: LangChain/CrewAI integration
pip install crewai langchain-core

Rust CLI + Library

# Prerequisites: Rust 1.70+ via rustup
curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh
source "$HOME/.cargo/env"
rustc --version  # Must be 1.70 or higher

# Build release binary
cargo build --release

# Install as CLI tool
cargo install --path .

Go SDK

# Prerequisites: Go 1.21+
# macOS: brew install go
# Linux: sudo apt install golang-go
# Windows: download from https://go.dev/dl/

go version  # Must be 1.21 or higher

# Download module dependencies
cd sdk/go && go mod tidy

# Run tests
go test -v

JavaScript SDK

# Prerequisites: Node.js 18+
node --version  # Must be 18 or higher

# SDK is built-in — no npm install needed
# Usage: const { AbirGuard } = require('./src/abir_guard');

Docker

# Prerequisites: Docker Engine 20.10+
docker --version

# Build and run
docker build -t abir-guard:latest .
docker run -d -p 9090:9090 -e ABIR_GUARD_API_KEY="your-key" abir-guard:latest

---

Quick Start

Install from Package Managers (Recommended)

# Python (PyPI)
pip install abir-guard

# Rust (crates.io)
cargo add abir_guard

# Go (GitHub)
go get github.com/Abiress/abir-guard/sdk/go

From Source

# Clone and install all components
git clone https://github.com/Abiress/abir-guard.git
cd abir-guard

# Python: install and test
pip install -e ".[dev]"
python3 tests/run_tests.py

# Rust: build and test
cargo build --release && cargo test

# Go: test
cd sdk/go && go test -v && cd ../..

# JavaScript: verify (Node.js)
node -e "const { AbirGuard } = require('./src/abir_guard'); new AbirGuard().generateKeyPair('test').then(console.log)"

---

Python SDK Guide

Basic Vault Operations

from abir_guard import Vault

vault = Vault()

# Generate keypair for an agent
pub, sec = vault.generate_keypair("finance_agent")

# Encrypt sensitive data
ct = vault.store("finance_agent", b"API_KEY=sk-abc123xyz")

# Decrypt data
plaintext = vault.retrieve("finance_agent", ct)
# → b"API_KEY=sk-abc123xyz"

# Auto-generate keys on first use
ct = vault.store("new_agent", b"data")  # keypair created automatically

# List and delete keys
vault.list_keypairs()        # ['finance_agent', 'new_agent']
vault.remove_keypair("new_agent")

Phase 3 Features

# Key Revocation (CRL)
from abir_guard.revocation import RevocationList, RevocationReason
crl = RevocationList()
crl.revoke("compromised-key", RevocationReason.COMPROMISED, "admin", "Key leaked")
crl.is_revoked("compromised-key")  # True

# Automatic Key Rotation
from abir_guard.rotation import KeyRotationManager
mgr = KeyRotationManager(default_max_operations=1000)
mgr.register_key("agent-1", max_operations=500)
mgr.record_usage("agent-1", "encrypt")
mgr.needs_rotation("agent-1")  # False (under limit)

# FIPS 140-3 Compliance Mode
from abir_guard.fips_mode import FIPSEncryptor
fips = FIPSEncryptor()
encrypted = fips.encrypt(data, key)
decrypted = fips.decrypt(ct, tag, nonce, key)

# Differential Privacy Entropy
from abir_guard.differential_privacy import DifferentialEntropyCollector
collector = DifferentialEntropyCollector(epsilon=0.5, sample_count=20)
entropy = collector.collect()  # 32 bytes of noise-injected entropy

# Remote Attestation
from abir_guard.attestation import IntegrityProof, AttestationVerifier
proof = IntegrityProof()
proof.compute(challenge="abc123")
verifier = AttestationVerifier()
verifier.verify_proof(proof.to_dict())  # True if untampered

MCP HTTP Server

from abir_guard.mcp_http import McpHttpServer

server = McpHttpServer(port=9090, api_key="your-secret-key", rate_limit=100)
server.start()

# curl http://localhost:9090/health
# curl -X POST http://localhost:9090 -H "Authorization: Bearer your-secret-key" ...

---

Rust CLI & Library Guide

CLI Commands

# Initialize vault with passphrase
./target/release/abir-guard -k "my-passphrase" init my-agent

# Encrypt / decrypt
./target/release/abir-guard -k "my-passphrase" encrypt my-agent "secret data"
./target/release/abir-guard -k "my-passphrase" decrypt my-agent "<ciphertext>" "<nonce>"

# SHAMIR secret sharing
./target/release/abir-guard shamir-split "my-passphrase" -t 3 -n 5
./target/release/abir-guard shamir-join "1:..." "3:..." "5:..."

# ML-DSA signatures
./target/release/abir-guard -k "my-passphrase" mldsa-init --key-id agent
./target/release/abir-guard -k "my-passphrase" mldsa-sign agent "data"

# Start MCP server
./target/release/abir-guard mcp-server --mode stdio

Library Usage

use abir_guard::Vault;

let vault = Vault::new();
let ct = vault.store(b"agent-1", b"secret data").unwrap();
let plain = vault.retrieve(b"agent-1", &ct).unwrap();
assert_eq!(plain, b"secret data");

Persistent vault with passphrase:

use abir_guard::persistent_vault;

let vault = persistent_vault::get_vault("my-passphrase");
let ct = persistent_vault::store_encrypted(&vault, "agent", b"secret", "my-passphrase").unwrap();

---

Go SDK Guide

import "github.com/abir-guard/abir-guard/sdk/go"

vault := abirguard.NewVault()

// Generate keypair
vault.GenerateKeypair("agent-1")

// Encrypt / decrypt
ct, _ := vault.Encrypt("agent-1", []byte("sensitive data"))
plain, _ := vault.Decrypt("agent-1", ct)

// Revoke key
vault.RevokeKey("compromised", "compromised", "admin", "Key leaked")

// Rotate key
vault.RotateKey("agent-1")

// Check rotation status
meta, _ := vault.GetMetadata("agent-1")
fmt.Printf("Operations: %d encrypt, %d decrypt\n", meta.EncryptCount, meta.DecryptCount)

// Audit log
for _, entry := range vault.GetAuditLog() {
    fmt.Printf("[%s] %s: %s\n", entry.Timestamp, entry.Action, entry.KeyID)
}

---

JavaScript SDK Guide

const { AbirGuard, AbirGuardMCP } = require('./src/abir_guard');

const vault = new AbirGuard();

const { publicKey, secretKey } = await vault.generateKeyPair('agent-1');
const { ciphertext, nonce, authTag } = await vault.encrypt('agent-1', 'API_KEY=sk-...');
const plaintext = await vault.decrypt('agent-1', { ciphertext, nonce, authTag });

// Rotate key (kill switch)
await vault.rotateKey('agent-1');

// MCP client
const mcp = new AbirGuardMCP(9090);
const result = await mcp.encrypt('agent-1', 'secret data');

---

MCP Server Guide

JSON-RPC Methods

Method	Params	Response	Description
generate_key	{key_id}	{key_id, generated: true}	Create keypair
encrypt	{key_id, data}	{nonce, ciphertext, key_id}	Encrypt data
decrypt	{key_id, ciphertext}	{plaintext}	Decrypt data
list_keys	{}	{keys: [...]}	List active keys
delete_key	{key_id}	{deleted: true}	Remove keypair
add_canary	{}	{canary_id}	Plant honeypot key
check_canary	{}	{breach_detected: bool}	Check for breaches
audit_log	{limit}	{entries: [...]}	View audit log
clear_cache	{}	{cleared: true}	Clear memory cache
info	{}	{name, version, mcp_version}	Server info

HTTP Endpoints

Endpoint	Auth	Description
POST /	Bearer token	MCP JSON-RPC gateway
GET /health	Public	Health check
GET /audit	Bearer token	Last 100 audit entries

---

LangChain & CrewAI Integration

LangChain

from abir_guard.langchain import get_langchain_tools

tools = get_langchain_tools()
# [SilentQKeyGenTool, SilentQEncryptTool, SilentQDecryptTool]

CrewAI

from abir_guard.crewai import get_crewai_tools

tools = get_crewai_tools()
# [KeyGenCrewTool, EncryptCrewTool, DecryptCrewTool]

---

Docker Deployment

# Build image
docker build -t abir-guard:latest .

# Run with API key and persistent volume
docker run -d --name abir-guard \
  -p 9090:9090 \
  -e ABIR_GUARD_API_KEY="your-secret-key" \
  -v abir-keys:/root/.abir_guard \
  abir-guard:latest

# Health check
curl http://localhost:9090/health

# Encrypt via HTTP
curl -X POST http://localhost:9090 \
  -H "Authorization: Bearer your-secret-key" \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","id":1,"method":"encrypt","params":{"key_id":"agent","data":"secret"}}'

---

HSM & TPM Integration

from abir_guard.abir_hsm import HSMKeyStore, TPMKeyStore

# Auto-detect best backend per OS
hsm = HSMKeyStore()
# macOS → Keychain, Windows → Credential Manager, Linux → file/secret_service

hsm.store_secret("my-api-key", b"sk-abc123")
secret = hsm.retrieve_secret("my-api-key")

# TPM 2.0 hardware detection
tpm = TPMKeyStore()
if tpm.is_available():
    print("TPM hardware detected — keys can be hardware-sealed")

Phase 2 Hardware Security Features

YubiKey / FIDO2 Integration

from abir_guard import YubiKeyManager

yk = YubiKeyManager()

# Generate hardware-backed key
cred_id = yk.generate_key("agent-1", "ed25519")

# Sign data (requires YubiKey touch in production)
signature = yk.sign("agent-1", b"data to sign")

# Encrypt/decrypt with YubiKey-backed keys
ct, nonce = yk.encrypt_with_yubikey("agent-1", b"secret")
plaintext = yk.decrypt_with_yubikey("agent-1", ct, nonce)

TPM 2.0 Seal/Unseal

from abir_guard import TPM2Sealer

tpm = TPM2Sealer()

# Seal data to TPM PCR values (hardware-bound)
sealed = tpm.seal(b"master-key", pcr_indices=[0, 7])

# Unseal - only works if system state matches
recovered = tpm.unseal(sealed)

Hardware Enclave Detection

from abir_guard import HardwareEnclave

enc = HardwareEnclave()
print(f"Platform: {enc.platform}")
print(f"Available: {enc.is_available()}")

# Generate hardware-backed key
enc.generate_key("agent-1")

# Seal/unseal using best available hardware
sealed = enc.seal(b"secret", "agent-1")
recovered = enc.unseal(sealed, "agent-1")

# Get attestation report
report = enc.attest(b"challenge-nonce")

---

Quantum Readiness

What "Quantum-Ready" Means for Abir-Guard

Threat	Mitigation	Status
Harvest Now, Decrypt Later	ML-KEM-1024 key encapsulation (NIST FIPS 203)	Production Ready
Quantum Key Extraction	AES-256-GCM with 256-bit keys (Grover-resistant)	Production
Signature Forgery (Shor's)	ML-DSA-65 digital signatures (NIST FIPS 204)	Production
Side-Channel Quantum Attacks	Differential privacy entropy + constant-time comparison	Production
Memory Scraping	Zero-copy memory policy + explicit key zeroization	Production
Future Quantum Break	Hybrid KEM (ML-KEM + X25519) — both must break	Production

Current Quantum Status

• AES-256-GCM: Quantum-safe. Grover's algorithm reduces effective strength to 128-bit, still secure.
• ML-DSA-65: Post-quantum signatures deployed and tested. 3309-byte signatures, constant-time operations.
• ML-KEM-1024: Production-ready. Implemented in Python via pqcrypto (PQClean-backed) and Rust via ml-kem crate (pure Rust). Full keygen, encapsulation, and decapsulation roundtrip verified.
• SHAMIR + Argon2id: Classical but quantum-safe for their use cases (threshold sharing, key derivation).

Mission Alignment 🇮🇳🌍

This project aligns with and supports:

• 🇮🇳 Indian Quantum Mission — India's National Quantum Mission (NQM) aims to develop quantum technologies for communication, computing, and sensing. Abir-Guard provides NIST-standard post-quantum cryptography to safeguard India's quantum infrastructure against Harvest Now, Decrypt Later threats.
• 🌍 Global Quantum Mission — Aligns with the worldwide transition to post-quantum cryptography as mandated by NIST, ENISA, and national cybersecurity agencies. Abir-Guard implements NIST FIPS 203 (ML-KEM) and FIPS 204 (ML-DSA) for quantum-resilient data protection.
• 🇮🇳🌍 Indian AI Mission — Supports India's AI sovereignty initiative by providing quantum-secure memory vaults for AI agents, ensuring API keys, model weights, and agent memory remain protected against future quantum attacks. Built in India, for the world.

After Quantum Breakthrough

1. All ML-KEM-1024 backends are production-ready — no additional setup needed
2. Python uses pqcrypto (PQClean-backed) for native ML-KEM-1024
3. Rust uses ml-kem crate (pure Rust, zero dependencies)
4. Existing hybrid keys remain valid during transition

---

Security Architecture

Hybrid KEM Design

┌──────────────────────────────────────────────────────────┐
│              Hybrid Key Encapsulation                     │
│  ML-KEM-1024 (PQC)  +  X25519 (Classical ECDH)          │
│  Security: Both must be broken to compromise              │
└──────────────────────────────────────────────────────────┘
                          │
                          ▼
┌──────────────────────────────────────────────────────────┐
│              Envelope Encryption                         │
│  AES-256-GCM (NIST FIPS 197)                            │
│  96-bit random nonce + 128-bit auth tag per message     │
└──────────────────────────────────────────────────────────┘

Defense-in-Depth Layers

Layer	Controls
Cryptography	AES-256-GCM, ML-KEM-1024, ML-DSA-65, Argon2id, HKDF-SHA256
Memory Safety	Zero-copy policy, explicit key zeroization, Rust ownership model
Network	Bearer token auth, rate limiting (100/min), TLS support, localhost default
Integrity	SHA-256 hash-chain audit logs, HMAC-signed CRL, tamper-evident vault
Runtime	Remote attestation, canary honeypots, Spectre/Meltdown noise injection
Lifecycle	Auto key rotation (time/usage), revocation, expiry policies
Compliance	FIPS 140-3 strict mode, approved algorithms only, audit trail

---

Run Tests

# Full test suite (recommended before deployment)
cargo build --release && cargo test && \
python3 tests/run_tests.py && \
pytest tests/test_abir_guard.py tests/test_phase3.py -v && \
cd sdk/go && go test -v && cd ../..

# Individual test suites
cargo test                          # Rust: 32 tests
pytest tests/test_abir_guard.py -v  # Python Phase 1: 17 tests
pytest tests/test_phase3.py -v      # Python Phase 3: 24 tests
pytest tests/test_phase2_hardware.py -v  # Python Phase 2: 24 tests
cd sdk/go && go test -v             # Go: 12 tests
python3 tests/run_tests.py          # Manual suites: 5/5

All 109 tests pass across Rust, Python, and Go.

---

Project Structure

abir_guard/
├── abir_guard/              # Python package (15 modules)
│   ├── __init__.py          # Core Vault, HybridEncryptor, McpServer, AuditLogger
│   ├── ml_kem.py            # ML-KEM-1024 + X25519 hybrid KEM (real ECDH)
│   ├── yubikey_integration.py # YubiKey/FIDO2 integration (software fallback)
│   ├── tpm2_seal.py         # TPM 2.0 seal/unseal (tpm2-tools CLI)
│   ├── hardware_enclave.py  # Apple SE, Intel SGX, AMD SEV detection
│   ├── langchain.py         # LangChain tool integration (3 tools)
│   ├── crewai.py            # CrewAI tool integration (version-compatible)
│   ├── abir_hsm.py          # HSM/TPM integration (Keychain, CredMgr, file, TPM)
│   ├── mcp_http.py          # Hardened HTTP MCP server (auth, rate limit, TLS)
│   ├── crypto_store.py      # Encrypted disk persistence (Argon2id + AES-GCM + HMAC)
│   ├── revocation.py        # CRL-style key revocation with HMAC signing
│   ├── rotation.py          # Automatic key rotation (time-based + usage-based)
│   ├── fips_mode.py         # FIPS 140-3 compliance mode (strict NIST algorithms)
│   ├── differential_privacy.py # Laplace noise entropy (Spectre/Meltdown defense)
│   └── attestation.py       # Remote attestation (runtime integrity verification)
├── src/                     # Rust source (12 modules)
│   ├── lib.rs               # Library entry point + re-exports
│   ├── main.rs              # CLI binary (clap subcommands, passphrase, validation)
│   ├── quantum_kernel.rs    # Hybrid encryption + 200ms watchdog + zeroization
│   ├── entropy_inject.rs    # CPU jitter entropy collector
│   ├── zero_copy.rs         # Zero-copy vault with LRU-encrypted cache
│   ├── mcp_gateway.rs       # MCP JSON-RPC server (10 methods)
│   ├── persistent_vault.rs  # Encrypted file persistence (Argon2id + AES-GCM + ML-DSA)
│   ├── kdf.rs               # Argon2id key derivation (OWASP: 64MB, 3 iter)
│   ├── shamir.rs            # SHAMIR Secret Sharing (t, n) over GF(251)
│   ├── ml_dsa.rs            # ML-DSA-65 signatures (NIST FIPS 204)
│   ├── revocation.rs        # Key revocation/blacklist (CRL, HMAC-signed)
│   ├── rotation.rs          # Automatic key rotation manager
│   └── differential_privacy.rs # Laplace noise + Spectre/Meltdown defender
├── sdk/
│   ├── go/                  # Go SDK (AES-256-GCM vault with CRL, rotation, metadata)
│   │   ├── abirguard.go     # Core implementation
│   │   ├── abirguard_test.go # 12 unit tests
│   │   └── go.mod           # Module definition
│   └── js/                  # JavaScript SDK (Node.js crypto + MCP client)
│       └── abir_guard.js    # Basic vault + MCP client
├── examples/                # Usage examples
├── tests/                   # Test suites (Python)
│   ├── run_tests.py         # Manual test runner (5 suites)
│   ├── test_abir_guard.py   # Pytest Phase 1 (17 tests)
│   ├── test_phase2_hardware.py # Pytest Phase 2 (24 tests)
│   └── test_phase3.py       # Pytest Phase 3 (24 tests)
├── scripts/                 # Publishing and debugging scripts
│   ├── publish-pypi.sh      # PyPI publishing script
│   ├── publish-crates.sh    # crates.io publishing script
│   └── debug.sh             # Full project debug & verification
├── Cargo.toml               # Rust dependencies (edition 2021)
├── pyproject.toml           # Python package config (v3.1.0)
├── PUBLISHING.md            # PyPI and crates.io publishing guide
├── Dockerfile               # Container build (hardened MCP server)
├── LICENSE                  # MIT License (2026)
├── README.md                # This file
├── THREAT_MODEL.md          # Zero-trust threat model
├── SECURITY.md              # Vulnerability reporting
├── CONTRIBUTING.md          # Contribution guidelines
├── CODE_OF_CONDUCT.md       # Community standards
├── CITATION.cff             # Academic citation
└── TASKS.md                 # Feature status and roadmap

---

Roadmap

Phase 1: Bedrock (Complete)

• X25519 hybrid KEM with AES-256-GCM
• Memory zeroization (Rust zeroize)
• Security Watchdog (200ms)
• Encrypted disk persistence
• Input validation
• MCP JSON-RPC Gateway
• Python + Rust + JavaScript SDKs
• LangChain + CrewAI integration
• HSM + TPM integration
• Docker + CI/CD
• Audit logging + canary keys

Phase 2: Hardware & Security (Complete)

• ML-DSA-65 signatures (NIST FIPS 204)
• SHAMIR secret sharing (GF(251))
• Argon2id KDF in Rust
• Real ML-KEM-1024 (Python: pqcrypto + Rust: ml-kem crate)
• YubiKey / FIDO2 integration (software fallback ready)
• TPM 2.0 seal/unseal (via tpm2-tools CLI)
• Apple Secure Enclave / Intel SGX / AMD SEV platform detection

Phase 3: Ecosystem & Hardening (Complete)

• Key revocation (CRL, HMAC-signed)
• Automatic key rotation (time/usage)
• FIPS 140-3 compliance mode
• Differential privacy entropy
• Remote attestation
• Go SDK
• PyPI publishing (pip install abir-guard)
• crates.io publishing (cargo add abir_guard)

---

Upcoming Phases

🚀 Phase 4: Enterprise & Cloud Integration (Q1 2026)

Production readiness for enterprise deployments and cloud-native workflows

• Real YubiKey/FIDO2 Hardware — FIDO2/CTAP2 operations, touch confirmation, PIV slot management
• Native TPM 2.0 API — tpm2-tss library integration, PCR policy automation
• AWS KMS / GCP KMS Integration — Cloud KMS backends, envelope encryption
• HashiCorp Vault Integration — Vault transit engine backend, enterprise secret management
• Kubernetes Operator — Auto-inject vault sidecars, secret rotation, Helm charts
• Multi-Tenant Support — Organization/workspace isolation, RBAC, audit partitioning
• Performance Benchmarking — Async I/O, connection pooling, 10k ops/sec target
• OpenTelemetry Integration — Metrics, traces, distributed tracing for vault operations

🔐 Phase 5: Advanced AI Security & Compliance (Q2 2026)

AI-specific security patterns, regulatory compliance, multi-agent workflows

• Complete JavaScript SDK — ML-KEM-1024, ML-DSA-65, WebCrypto API, browser extensions
• Model Weight Encryption — Encrypt LLM weights at rest, secure fine-tuning pipelines
• Prompt Injection Shield — Detect/encrypt malicious prompts, prompt signature verification
• GDPR/CCPA/HIPAA Compliance — Data retention policies, right-to-erasure, audit exports
• Multi-Agent Key Sharing — Threshold encryption for agent swarms, quorum-based access
• Secure Enclave for LLMs — TEE-based inference (Intel TDX, AMD SEV-SNP), attested compute
• Zero-Knowledge Proofs — Prove encryption without revealing data, compliance audits
• AI Red-Teaming Tools — Automated attack simulation, breach scenario testing

🌐 Phase 6: Distributed & Quantum Ecosystem (Q3 2026)

Distributed vault architecture, quantum network readiness, ecosystem expansion

• Federated Vault Network — Distributed vault mesh, CRDT-based sync, conflict resolution
• Quantum Key Distribution (QKD) — QKD network integration, BB84 protocol support
• Post-Quantum TLS — Hybrid TLS 1.3 with ML-KEM-1024, secure transport layer
• WASM Compilation — Browser-native vault, edge computing, Deno/Cloudflare Workers
• Apple Secure Enclave Native — Swift bindings, native SE API, macOS/iOS SDK
• Intel SGX Enclave — Actual enclave creation, remote attestation, secure compute
• Decentralized Identity (DID) — W3C DID integration, self-sovereign identity, verifiable credentials
• HSM Cluster — Multi-HSM load balancing, failover, geographic distribution

Contributing

See CONTRIBUTING.md for guidelines, coding standards, and the PR checklist. I welcome contributions from developers, security researchers, and AI engineers.

---

Project Governance

Document	Purpose
THREAT_MODEL.md	Zero-trust threat model, trust boundaries, mitigations
SECURITY.md	Vulnerability reporting policy, disclosure process
CONTRIBUTING.md	Contribution guidelines, code style, PR checklist
PUBLISHING.md	PyPI and crates.io publishing guide
CODE_OF_CONDUCT.md	Community standards and enforcement
CITATION.cff	Academic citation for research papers

---

License

MIT License. See LICENSE for details.

Copyright (c) 2026 Abir Maheshwari

---

Developer

Abir Maheshwari
Founder at Artificial Quantum Dyson Intelligence, Biro Labs, Aquilldriver
AI Engineer | Quantum Computing Researcher

Connect

• Email: [e-mail]
• LinkedIn: https://in.linkedin.com/in/abirmaheshwari
• Instagram: @anantraga31
• Medium: https://office.qz.com/@abirmaheshwari

---

Built with Rust, Python, Go, JavaScript · Secured by NIST PQC, AES-256-GCM, Argon2id, ML-DSA-65, ML-KEM-1024 · Licensed under MIT 2026

---

🇮🇳🌍 Mission Support

Mission	Badge	Description
🇮🇳 Indian Quantum Mission		Quantum-resilient cryptography for India's National Quantum Mission
🌍 Global Quantum Mission		NIST FIPS 203/204 compliant worldwide
🇮🇳🌍 Indian AI Mission		Quantum-secure memory vaults for sovereign AI agents

🇮🇳 Made in India, for the World.