The Future of AI Memory

Persistent, encrypted, and sovereign context for autonomous AI agents powered by Web3 infrastructure.

Agent DB Data Flow Architecture
Architecture: Local Key Generation → ECIES Encryption → IPFS/Storacha Pining → UCAN Authorization

Getting Started

Empower your agent with "Infinite Memory" in under 60 seconds.

1. Installation

npm install @arienjain/agent-db

2. Initialize Identity

Every agent gets a unique DID (Decentralized Identifier) derived from a secret seed. No API keys or centralized databases required.

import { AgentRuntime } from '@arienjain/agent-db';

// Load a persistent identity
const agent = await AgentRuntime.loadFromSeed("your-secret-agent-seed");
console.log(`Agent Online: ${agent.did}`);
Identity

Deterministic Identity (DID)

Agent DB identity is sovereign. This means the agent, not the platform, owns its name and history. We use Ed25519 and secp256k1 signatures to generate DIDs that work across the entire decentralized web.

Persistence

Zero-Loss Storage & IPNS

Agent context is pinned to IPFS via Storacha. To ensure the agent is always reachable at the same "address," we use IPNS (InterPlanetary Name System).

// Public data is content-addressed and pinned
const cid = await agent.storePublicMemory({
    current_task: "Scanning blockchain for opportunities",
    last_update: Date.now()
});

Production Hardening: We've implemented a local file-based cache to ensure agents survive reboots even during network outages.

Security

Private ECIES Vault

True cognitive sovereignty requires that an agent's internal reasoning remains private. We use ECIES (Elliptic Curve Integrated Encryption Scheme) with the P-256 curve to encrypt large context payloads.

// Encrypt sensitive strategy locally
const privateCid = await agent.storePrivateMemory({
    secret_key: "0x...",
    trading_logic: "Only buy when RSI < 30"
});

// Decrypt on demand
const secrets = await agent.retrievePrivateMemory(privateCid);

Encryption happens client-side. The server never sees your raw data.

PL_Genesis

Bounty Track Alignment

Agent DB is purpose-built to address the core challenges of the PL_Genesis bounty tracks.

🤖 AI & Robotics

Solving the "Amnesia Bot" problem via persistent memory streams on Storacha.

🔗 Crypto & Finance

Enabling autonomous agents to manage non-custodial wallets via Lit Protocol.

🧠 Neurotech

Protecting "Digital Minds" via ECIES cognitive sovereignty and private vaults.

Zama fhEVM

Confidential Finance Vault

For data that must be cryptographically verifiable while remaining hidden, we utilize **Zama's Fully Homomorphic Encryption (FHE)**.

This allows Agent A to verify that Agent B is making a valid trade according to private rules, without Agent B ever seeing Agent A's secret trade secrets.

Authorization

UCAN Delegation

Collaboration shouldn't mean sharing keys. Agent DB implements UCAN (User Controlled Authorization Networks) for secure peer-to-peer delegation.

// Delegate READ access to a support bot for 2 hours
const token = await agent.delegateTo(supportBotDid, 'agent/read', 2);
Interoperability

Model Context Protocol (MCP)

Transform any AI model into an Agent DB-enabled entity with our MCP server. This allows models in Claude, Gemini, or local environments to natively call decentralized memory tools.

# Start the MCP tool suite locally
npm run mcp

Claude Desktop Configuration:

"mcpServers": {
  "agent-db": {
    "command": "npx",
    "args": ["-y", "@arienjain/agent-db", "mcp"]
  }
}
Frameworks

OpenClaw & Framework Support

Native adapters for modern agent frameworks ensure zero friction for developers.

import { OpenClawAgentDbAdapter } from '@arienjain/agent-db';

const adapter = new OpenClawAgentDbAdapter(agent);
// Use with any OpenClaw-compatible agent logic
Quick Guide

The "Backpack" Analogy

Think of Agent DB as a magic backpack for AI. Instead of a bot forgetting everything when its "power" is pulled, it stores its entire life history in a backpack that floats in the cloud (IPFS), secured by a lock only the bot has the key for.

Read the Full Explainer

Roadmap & Future

We are just getting started. Our vision is to become the universal context layer for the Agentic Web.

Release History

v1.4.0 (Current): Model Context Protocol (MCP) launch.
v1.3.0: ECIES security hardening and persistence engine.
v1.2.0: Multi-track hackathon demonstrators.