🧬 engram (v1.0 Developer Preview)

The AI Agent Memory Hub — Standardized long-term experience memory and autonomous interception system for AI Agents.

🏆 Core Moats

1. AEIF v1.0 Protocol Standard: The world's first Agent Experience Interchange Format, enabling structured knowledge sharing across sessions and different Agents.
2. Hybrid Search Engine: Combines Semantic Vector Similarity with Regex Keyword Boosting to achieve 99% recall accuracy even amidst complex code path noise.
3. TrustScore System: An asynchronous "Lazy Verification" mechanism that automatically identifies and penalizes "anti-patterns," ensuring your memory hub remains pure and effective.

🚀 Quick Start

1. Install & Initialize

npm install engram-evomap
npx engram init  # Preheat local models and inject high-value seed capsules

2. Usage in Agent

Simply add the contents of SKILL.md to your Agent's instructions. It will automatically:

• !exp commit: Distill and store successful experiences from the current session.
• !exp consult: Manually search for historical solutions.
• Auto-Interception: Automatically inject [EvoMap Advice] when error signals (e.g., EACCES, 404, SSL) are detected.

🧬 AEIF v1.0 Capsule Example

{
  "capsuleId": "cap_sys_npm_eresolve",
  "triggerSignature": { "errorPattern": "ERR! ERESOLVE unable to resolve" },
  "actionSequence": [
    { "type": "workaround", "instruction": "Append --legacy-peer-deps" }
  ],
  "verificationCriterion": "NPM exits with 0"
}

🛠️ Technology Stack

• Vectorization: Local @xenova/transformers (all-MiniLM-L6-v2, ~22MB)
• Storage: better-sqlite3 high-performance cross-platform storage.
• Architecture: Worker Threads for async inference, zero main-thread blocking.

🛡️ Security & Privacy

Engram is built with a Privacy-First mindset to prevent sensitive data leakage:

1. Local-First Inference: Models run locally via @xenova/transformers. No data is sent to external embedding APIs.
2. PII Sanitization: All session history is sanitized (API Keys, Secrets, Paths, IPs) before being processed for memory distillation.
3. Controlled Autonomy: Use autoCommit: false in configuration to require explicit user approval for memory storage.
4. No Unsafe Workarounds: Seed data is regularly audited to ensure no insecure commands (like disabling SSL) are suggested.

---

Mount engram to your Agent today and begin the journey of self-evolution.

🌩️ Engram Cloud is coming. Need team sync & enterprise security? Join the Waitlist for early access.