Continuo Memory System

Persistent memory and hierarchical compression for development environments

Overview

Continuo is a persistent memory system that provides semantic search and storage capabilities for development workflows. By separating reasoning (LLM) from long-term memory (Vector DB + hierarchical compression), the system maintains knowledge indefinitely, circumventing context window limitations.

Key Features

• Persistent Memory - Store and retrieve development knowledge across sessions

• Semantic Search - Find relevant information using natural language queries

• Hierarchical Compression - N0 (chunks) → N1 (summaries) → N2 (meta-summaries)

• MCP Integration - Seamless integration with IDEs via Model Context Protocol

• Cost Effective - 100% local (free) or hybrid (low-cost) deployment options

• FastMCP - Built on the modern MCP server framework

Quick Start

Installation

Usage

1. Start the memory server:

2. Configure your IDE (Qoder/Cursor):

Create .qoder/mcp.json (or .cursor/mcp.json):

3. Use in your IDE:

Architecture

Components

• Memory Server - ChromaDB + sentence-transformers for embeddings

• MCP Adapter - FastMCP server exposing search_memory and store_memory tools

• Hierarchical Compression - Multi-level context optimization (N0/N1/N2)

• Autonomous Mode - Optional automation with Observe → Plan → Act → Reflect cycle

Configuration

Local Embeddings (Free)

OpenAI Embeddings (Low-cost)

API

Tools

search_memory(query: str, top_k: int = 5, level: str | None = None) -> str

• Semantic search in persistent memory

• Returns relevant documents with similarity scores

store_memory(text: str, metadata: dict | None = None, level: str = "N0") -> str

• Store content in persistent memory

• Supports metadata tagging and hierarchical levels

get_memory_stats() -> str

• Get memory statistics (total documents, levels, etc.)

Hierarchical Levels

• N0 - Raw chunks (code snippets, conversations)

• N1 - Micro-summaries (5-10 chunks compressed)

• N2 - Meta-summaries (5-10 summaries compressed)

Examples

See the examples/memory/ directory:

• basic_usage.py - Simple store/retrieve operations

• hierarchical_demo.py - Multi-level compression examples

• auto_mode_demo.py - Autonomous mode demonstration

Documentation

• Setup Guide - Detailed installation instructions

• Architecture Specification - Complete technical documentation

• Code of Conduct - Community guidelines

Technology Stack

• Python 3.9+ - Core implementation

• ChromaDB - Vector database for embeddings

• Sentence Transformers - Local embedding generation (all-MiniLM-L6-v2)

• FastMCP - MCP server framework

• Model Context Protocol - IDE integration standard

Cost & Licensing

Embedding Providers

Software License

See COMMERCIAL_LICENSE.md for details.

Contributing

Contributions are welcome! Please read CONTRIBUTING.md for guidelines.

License

Continuo Memory System is dual-licensed:

📖 Open Source (AGPL v3)

FREE for:

• ✅ Individual developers and researchers

• ✅ Non-profit organizations and educational institutions

• ✅ Companies with <$1M revenue AND <10 employees

• ✅ Development, testing, and evaluation

• ✅ Open source projects (AGPL-compatible)

Requirements: Share source code of modifications under AGPL v3

See LICENSE for full AGPL v3 terms.

💼 Commercial License

REQUIRED for:

• ❌ Companies with ≥$1M revenue OR ≥10 employees

• ❌ Proprietary/closed-source products

• ❌ SaaS offerings without source disclosure

Benefits:

• ✅ No AGPL copyleft obligations

• ✅ Proprietary use rights

• ✅ Priority support (optional)

• ✅ Custom deployment assistance (optional)

Pricing: From $2,500/year (Bronze) to custom Enterprise

See COMMERCIAL_LICENSE.md for pricing and details.

💡 Why AGPL + Commercial?

• Sustainable Development: Commercial users fund ongoing maintenance

• Open Source Protection: AGPL prevents proprietary forks

• Fair Use: Small teams and non-profits use free indefinitely

• Community First: Core features always open source

Contact: licensing@continuo.dev (UPDATE) for commercial inquiries

Acknowledgments

Built using:

• Model Context Protocol - Protocol specification

• MCP Python SDK - MCP implementation

• ChromaDB - Vector database

• Sentence Transformers - Embedding models

Authors

• D.D. & Gustavo Porto

Note: This project implements the architecture described in continuo.markdown. For academic context and detailed specifications, refer to that document.