🚀 agentMemory

Hybrid Memory System for AI Coding Agents

Seamlessly integrate with KiloCode, Cline, and RooCode's built-in memory banks while providing powerful search, analytics, and automation.

---

🚀 Now Available as an Antigravity Skill!

agentMemory is now fully compatible with Antigravity. Use it as a skill to give your agents persistent, searchable memory that syncs with your project documentation.

See SKILL.md for usage instructions.

---

🎯 What Makes agentMemory Different?

KiloCode, Cline, and RooCode already have memory banks - but they're manual and limited.

Their Built-In Memory Banks

.kilocode/rules/memory-bank/     ← Markdown files
.clinerules/memory-bank/         ← Manually updated
.roo/memory-bank/                ← No search capability

The Problem with Native Memory Banks:

• ❌ Manual Maintenance: Agents must manually rewrite files to update memory.
• ❌ No Search: Agents must read entire files to find specific information.
• ❌ No Analytics: No way to track memory growth or patterns over time.
• ❌ Isolated: Memories are locked to a single project.

✅ The Solution: agentMemory

agentMemory fixes these limitations by upgrading your memory bank with:

.agentMemory/                    ← Our structured database
  ├── Bi-directional sync        ← Keeps their markdown updated
  ├── Powerful search            ← Query by type, tags, content
  ├── Visual dashboard           ← See trends and analytics
  └── MCP tools                  ← Programmatic access for AI

Benefits:

• ✅ Automatic sync - No manual "update memory bank" needed
• ✅ Searchable - Find memories by query, type, or tags
• ✅ Visual dashboard - Charts, trends, and analytics
• ✅ Cross-project - Query memories across all your projects
• ✅ Compatible - Works WITH their systems, not against them

---

🏗️ How It Works: Hybrid Architecture

Two-Layer System

┌─────────────────────────────────────────────────────────┐
│  Built-In Memory Banks (Markdown Files)                │
│  .kilocode/rules/memory-bank/                          │
│  .clinerules/memory-bank/                              │
│  .roo/memory-bank/                                      │
│  ├── Human-readable documentation                      │
│  ├── Git-committable                                   │
│  └── Read automatically by agents at session start     │
└──────────────┬──────────────────────────────────────────┘
               │
               │ ⬍ Bi-Directional Sync ⬊
               │
┌──────────────▼──────────────────────────────────────────┐
│  agentMemory (Structured Storage + MCP Server)         │
│  .agentMemory/                                         │
│  ├── JSON database for fast queries                   │
│  ├── MCP tools (memory_read, memory_search, etc.)     │
│  ├── Dashboard with analytics                         │
│  └── Cross-project search capabilities                │
└─────────────────────────────────────────────────────────┘

The Flow

1. Session Start - Passive Context Loading

AI reads markdown files (.kilocode/rules/memory-bank/)
  └─> Contains: Project context + MCP usage instructions
  └─> Sees: "This project uses agentMemory MCP tools"

2. During Session - Active Queries

AI needs to find auth patterns
  └─> Calls: memory_search({ query: "auth pattern" })
  └─> Gets: Structured results from our database
  └─> Faster and more accurate than file scanning

3. After Work - Automatic Sync

AI calls: memory_write({ key: "new-feature", ... })
  └─> We save to: .agentMemory/ (JSON database)
  └─> We sync to: .kilocode/rules/memory-bank/product.md
  └─> Next session: They see it in markdown automatically

---

✨ Key Features

🔄 Bi-Directional Sync

Markdown → MCP

• Parses existing memory bank files
• Imports sections as searchable memories
• Preserves human-friendly docs

MCP → Markdown

• Auto-generates markdown from MCP data
• Appends to appropriate files:
  • architecture.md ← Architecture decisions
  • systemPatterns.md ← Code patterns
  • techContext.md ← Tech stack choices
  • productContext.md ← Features
  • progress.md ← Status tracking

� Powerful Search

// Search across ALL memories
memory_search({ 
  query: "authentication", 
  type: "pattern",
  tags: ["security"]
})

// Returns structured results from our database
// Much faster than scanning markdown files

📊 Visual Dashboard

• Memory trends over time
• Agent activity tracking
• Type distribution (architecture vs patterns vs features)
• Recent changes timeline
• Cross-project insights

🤖 Multi-Agent Support

| Agent | Memory Bank Location | Sync Status | |-------|---------------------|-------------| | KiloCode | .kilocode/rules/memory-bank/ | ✅ Full sync | | Cline | .clinerules/memory-bank/ | ✅ Full sync | | RooCode | .roo/memory-bank/ | ✅ Full sync |

Files Synced:

• projectBrief.md / brief.md
• architecture.md / systemPatterns.md
• productContext.md / product.md
• techContext.md / tech.md
• activeContext.md / context.md
• progress.md
• decisionLog.md (RooCode)

---

📦 Installation

From Marketplace (1-Click)

1. Open VS Code
2. Extensions → Search "agentMemory"
3. Click Install
4. Reload VS Code

That's it! The extension will:

• ✅ Create MCP server configuration
• ✅ Inject memory-first instructions into memory banks
• ✅ Start bi-directional sync
• ✅ Enable dashboard

Manual Installation

git clone https://github.com/yourusername/agentMemory
cd agentMemory
npm install
npm run compile
npm run package
code --install-extension agentmemory-0.1.0.vsix

---

🎮 Usage

For End Users (Zero Config)

Once installed, agentMemory works automatically:

1. Import existing memory banks on first run
2. Sync new memories as agents create them
3. Provide MCP tools for fast queries
4. Update markdown files to keep agents in sync

No configuration needed!

For AI Agents (Automatic)

Agents will find mandatory instructions in their memory bank files:

# techContext.md

## agentMemory System (REQUIRED)

This project uses agentMemory for knowledge management.

### Required Workflow

1. Before ANY work: Call memory_search() to check existing knowledge
2. After ANY work: Call memory_write() to document decisions
3. Use memory_read() for specific pattern retrieval

Agents treat this as project architecture and follow it automatically.

---

🛠️ Memory Tools (MCP)

| Tool | Purpose | Example | |------|---------|---------| | memory_write | Save new memory | Document architecture decisions | | memory_read | Get specific memory | Retrieve OAuth implementation | | memory_search | Query by content/tags | Find all auth-related patterns | | memory_list | List by type | Show all architecture decisions | | memory_update | Modify existing | Append to existing pattern | | memory_stats | View analytics | Usage statistics |

---

� Project Structure

After installation, your project will have:

your-project/
├── .vscode/
│   └── settings.json           # MCP server config (auto-created)
│
├── .agentMemory/               # Our structured storage
│   ├── uuid-001.json          # Memory: OAuth architecture
│   ├── uuid-002.json          # Memory: API patterns
│   └── ...
│
├── .kilocode/rules/memory-bank/    # KiloCode memory bank
│   ├── brief.md                    # ⬍ Synced with our database
│   ├── architecture.md             # ⬍ Auto-updated
│   ├── product.md                  # ⬍ Auto-updated
│   └── tech.md                     # ⬍ Auto-updated
│
├── .clinerules/memory-bank/        # Cline memory bank
│   ├── projectBrief.md             # ⬍ Synced
│   ├── systemPatterns.md           # ⬍ Synced
│   └── ...                         # ⬍ Synced
│
└── .roo/memory-bank/               # RooCode memory bank
    ├── projectBrief.md             # ⬍ Synced
    ├── decisionLog.md              # ⬍ Synced
    └── ...                         # ⬍ Synced

All markdown files stay human-readable and git-friendly!

---

💡 Example Workflow

Day 1: Initial Implementation

User → KiloCode: "Create OAuth authentication"

1. KiloCode implements OAuth
2. Calls: memory_write({ key: "oauth-impl", type: "architecture", ... })
3. agentMemory saves to:
   • .agentMemory/uuid-001.json (our database)
   • .kilocode/rules/memory-bank/architecture.md (their markdown)

Day 3: Extension by Different Agent

User → Cline: "Add Google OAuth provider"

1. Cline reads .clinerules/memory-bank/architecture.md
2. Sees OAuth documentation (synced from our database)
3. Calls: memory_search({ query: "oauth" })
4. Gets structured results instantly
5. Implements Google provider consistently

Day 7: New Developer Onboarding

New Dev → RooCode: "How does authentication work?"

1. RooCode reads .roo/memory-bank/systemPatterns.md
2. Sees complete auth patterns (auto-synced)
3. Calls: memory_read({ key: "oauth-impl" })
4. Gets full OAuth architecture doc
5. ✅ Instant understanding - Zero learning curve

---

🎯 Why agentMemory?

| Feature | agentMemory | Built-In Banks | Standalone MCP | |---------|-------------|----------------|----------------| | Markdown Files | ✅ Synced | ✅ Manual | ❌ No | | Search | ✅ Fast indexed | ❌ No | ✅ Yes | | Analytics | ✅ Dashboard | ❌ No | ❌ No | | Automation | ✅ Auto-sync | ❌ Manual | ⚠️ Partial | | Multi-Agent | ✅ All 3 | ✅ Per-agent | ✅ All | | Git-Friendly | ✅ Yes | ✅ Yes | ⚠️ Depends | | **C