memory-layer 0.1.0

Persistent memory layer for AI coding assistants

memory-cli

A CLI memory layer for developers across editors (Cursor, VSCode, etc.) with prompt-driven memory mode.

🚀 New: Prompt-Driven Mode

The CLI now supports natural language queries with automatic memory recall!

# Ask any question - automatically recalls relevant context
memory "how did I integrate payments last week?"
memory "what was the bug I fixed yesterday?"
memory "show me my recent work on authentication"

# Check memory statistics
memory --stats

Installation

pip install .

Usage

🎯 Prompt-Driven Mode (Recommended)

Option 1: Direct natural language queries

# Using the installed command (if in PATH)
memory "how did I integrate payments last week?"

# Using Python directly
python prompt_direct.py "what was the bug I fixed yesterday?"

# Using batch file (Windows)
.\run_prompt.bat "show me my recent work on authentication"

Option 2: Using the chat command

memory chat "how did I integrate payments last week?"
memory chat --stats

🔧 Structured Commands (Legacy)

Option 1: Using the installed command (if in PATH)

memory init
memory add "Fixed login bug with JWT token"
memory recall "login"
memory attach cursor
memory ask "How do I fix a JWT login bug?"

Option 2: Using Python directly

python cli.py init
python cli.py add "Fixed login bug with JWT token"
python cli.py recall "login"
python cli.py attach cursor
python cli.py ask "How do I fix a JWT login bug?"

Option 3: Using the batch file (Windows)

.\run.bat init
.\run.bat add "Fixed login bug with JWT token"
.\run.bat recall "login"
.\run.bat attach cursor
.\run.bat ask "How do I fix a JWT login bug?"

Commands

Prompt-Driven Mode

• memory "your question" — Ask any question with automatic memory recall
• memory --stats — Show memory statistics
• memory chat "your question" — Alternative chat interface

Structured Commands

• init — Initializes a .memory folder with config and SQLite DB.
• add "note or context" — Stores the note with timestamp and project path.
• recall [query] — Searches for notes matching the query.
• attach editor — Stub for future editor integration.
• ask "prompt" — Ask the LLM with memory context.

LLM Integration

The tool supports multiple LLM providers. Add your API key to .memory/config.json:

OpenAI

{
  "llm_provider": "openai",
  "llm_api_key": "YOUR_OPENAI_KEY"
}

Anthropic

{
  "llm_provider": "anthropic",
  "llm_api_key": "YOUR_ANTHROPIC_KEY"
}

Google Gemini (Recommended)

{
  "llm_provider": "gemini",
  "gemini_api_key": "YOUR_GEMINI_KEY"
}

To get a Gemini API key:

1. Visit Google AI Studio
2. Create an API key
3. Update your config.json

Test your setup:

python test_gemini.py

How It Works

Prompt-Driven Mode

1. Natural Language Input: You ask any question in plain English
2. Automatic Memory Recall: System searches for relevant past conversations and notes
3. Context Enrichment: Your question is combined with relevant memory context
4. LLM Processing: Enhanced prompt is sent to OpenAI/Anthropic/Gemini
5. Automatic Storage: Both your question and the LLM response are saved for future reference

Database Schema

• notes table: Stores all conversations with embeddings support
• memory table: Legacy table for backward compatibility

Extensibility

The code is modular for future commands like sync, proxy, and editor integrations.

For best keyword extraction, run:

import nltk; nltk.download('punkt'); nltk.download('averaged_perceptron_tagger')