🧠 AndyAI Operational Memory

Semantic memory is not enough.
Operational Memory connects what happened, what is happening, what is trusted, and what should happen next.

What this repo is

AndyAI Operational Memory is a trust-aware memory layer for AI systems and agent workflows.

It is built to carry:

• durable continuity
• lifecycle-aware records
• trust and evidence metadata
• replayable artifacts
• release discipline

Core Formula

Memory + State + Direction + Trust + Action

Thesis

Semantic memory helps AI retrieve the past.
Operational memory helps AI work with continuity.

This repo pushes beyond retrieval by adding:

• lifecycle
• promotion
• evidence
• signatures
• replay
• release discipline

Quick Start

npm install
cp .env.example .env
npm run check
npm run trust:all
npm run release:all

Why it exists

Modern AI systems often fail because context is fragmented, temporary, and hard to verify.

This repo defines a practical foundation for systems that need more than chat history: they need durable, trusted, operational continuity.

---

Positioning

This is not:

• a chatbot
• a note-taking app
• a vector database demo

This is:

• a knowledge substrate
• a continuity engine
• a trust-aware memory layer
• a foundation for agent systems

---

Architecture

See:

• ARCHITECTURE.md
• MEMORY_LIFECYCLE.md
• RANKING_MODEL.md
• GRAPH_MODEL.md
• TRUST_MODEL.md
• docs/interoperability.md

Core Layers

1. Memory Layer

Stores structured memories with semantic retrieval support.

2. State Layer

Connects current operational context and active workflow state.

3. Direction Layer

Represents goals, plans, and trajectory.

4. Trust Layer

Adds evidence, authority, promotion flow, signatures, and verification.

5. Execution Layer

Supports integration with agents, tooling, automation, and release workflows.

---

Memory Model

Operational Memory requires more than similarity search.

It requires records that can evolve through time.

Minimal record concepts

• memory type
• status
• trust level
• authority level
• evidence
• related records
• replacement/supersession
• checksum/signature compatibility

Canonical types

• decision
• case
• pattern
• preference
• plan
• reference

Canonical lifecycle

• draft
• active
• verified
• superseded
• deprecated
• archived

---

Trust Model

Knowledge is not truth without evidence.

The trust layer adds:

• evidence links
• evidence hashes
• authority levels
• promotion gates
• signed exports
• verification flow
• replay-ready manifests

This moves memory from “helpful retrieval” toward “portable trust artifact.”

---

Implemented Foundations

This repository includes foundations for:

• Supabase + pgvector memory storage
• semantic retrieval
• ranking beyond cosine similarity
• duplicate blocking
• staged promotion flow
• audit logging
• signed trust bundles
• trust bundle verification
• replay manifest export
• CI trust gates
• release packaging

---

Project Structure

andyai-operational-memory/
├── README.md
├── ARCHITECTURE.md
├── MEMORY_SCHEMA.json
├── MEMORY_LIFECYCLE.md
├── RANKING_MODEL.md
├── GRAPH_MODEL.md
├── TRUST_MODEL.md
├── package.json
├── tsconfig.json
├── .env.example
├── sql/
├── src/
├── schemas/
├── scripts/
├── docs/
├── examples/
├── trust/
├── release/
├── dist/
└── 00_CREATE_REMOTE_FIRST.md

---

Trust Commands

Generate local signing keys:

mkdir -p .keys
npm run cli -- gen-keys .keys/memory-private.pem .keys/memory-public.pem

Export a signed trust bundle:

npm run export:bundle -- <memory-id> .keys/memory-private.pem exports

Verify a trust bundle:

npm run verify:bundle -- exports/memory-<id>.trust.json .keys/memory-public.pem

Export a replay manifest:

npm run replay:export -- <project-id> exports

---

Release Discipline

This repo includes CI trust gates for:

• TypeScript integrity
• trust bundle validation
• replay manifest validation
• release manifest generation
• trust report generation

Local trust run

npm run trust:all

One-command release packaging

npm run release:all

---

Intended Audience

This project is for:

• AI engineers
• agent builders
• infra/tooling teams
• trust/safety-minded developers
• advanced personal knowledge system builders
• teams building long-running AI workflows

---

Current Status

v1.0.2 — README Hero Polish

This release tightens the first-screen presentation while preserving the trust-aware architecture, lifecycle model, and release discipline established in v1.0.0 and v1.0.1.

---

Canonical Insight

Semantic Memory → finds
Operational Memory → carries
Trust Memory → proves

---

Final Insight

AI does not only need more intelligence.

It needs better memory infrastructure.