Problem and Design Goals

This page summarizes the product intent behind Engram SDK and how it shapes the technical design.

The Core Problem

Most AI agents are strong reasoners but weak operators:

• memory disappears across restarts and deployments
• they rely on humans for wallet management and funding
• actions are hard to verify externally
• multi-agent context handoff is usually centralized

Engram addresses this by making memory operations decentralized, autonomous, and verifiable.

Product Goals

Engram is designed around four concrete goals:

1. Persistent decentralized memory
   Agent state is stored as content-addressed data with explicit retention policy.

2. Autonomous economic execution
   Agents can execute storage operations without manual transaction handling in normal flow.

3. Confidential default path for sensitive state
   High-value contexts can be encrypted before storage and decrypted only under authorized conditions.

4. Cryptographic receipts for trust and auditability
   Operations can be anchored with on-chain attestations so external systems can verify what happened.

Design Principles

These goals are enforced with deliberate engineering constraints:

• Deterministic policy resolution (resolvePolicy) for predictable behavior
• Pre-flight safety checks (budget guard, balance checks, typed errors)
• Composable provider interface so runtime behavior stays stable across backends
• CID-first memory model so references are immutable and portable
• Recoverability-first indexing so agents can rebuild active memory state after disruption

What “Autonomous” Means in Practice

In Engram, autonomy does not mean “unbounded execution.” It means:

• agents can complete memory lifecycle operations with minimal operator intervention
• every irreversible operation has explicit checks and failure surfaces
• budgeting, attestation, and logging are first-class, not afterthoughts

What “Verifiable” Means in Practice

Verifiability comes from combined evidence:

• storage receipts with CID and spend metadata
• structured operation logs
• optional on-chain attestations for durable audit trails

This enables “agents with receipts” workflows where external parties can independently verify operation history.

Scope Boundaries

Engram is intentionally focused on memory infrastructure. It does not try to be:

• a full planning/execution agent framework
• an orchestration platform for every runtime concern
• a replacement for application-level authorization policy

Instead, it provides robust primitives that those systems can build on.