Skip to main content

​
Context Engineering, Not Information Retrieval

Jean Memory’s core philosophy is Context Engineering, not just Information Retrieval. This means the system doesn’t just store and retrieve memories—it intelligently engineers context for your AI assistant. This is the key to making AI truly personal and useful. The system is designed to:
  • Select the right information at the right time.
  • Synthesize insights from disparate memories.
  • Understand relationships between memories.
  • Predict what context will be most useful.
This is a continuous process. Memories are constantly being saved and analyzed in the background. When a query comes in, the system intelligently decides whether new context is required and, if so, what depth of search is necessary to provide the most relevant response.

​
The Orchestration Engine

The jean_memory API is the heart of the system. It’s the primary interface for your AI to interact with the memory layer, orchestrating various underlying functions to provide the right context at the right depth.

​
Speed and Depth Control

Jean Memory provides four distinct speed modes that allow you to balance response time with context depth:
  • Fast Mode: Direct memory search (0.5-1s) for real-time applications
  • Balanced Mode: AI synthesis with Gemini 2.5 Flash (3-5s) for conversational responses
  • Autonomous Mode: Intelligent orchestration with variable latency for complex decision-making
  • Comprehensive Mode: Deep document analysis (20-30s) for research tasks
The autonomous mode is particularly intelligent - it analyzes the conversation context to autonomously determine whether information should be saved, how much context to retrieve, and what depth of analysis is needed. While this may take longer than other modes for complex analysis, it prioritizes intelligent decision-making over consistent response times. Learn more about configuring speed modes →

​
Context Strategies

The orchestrator uses three primary strategies to provide the right context at the right time:
  1. Narrative Primer: For new conversations, the system retrieves a high-level user narrative to provide immediate, foundational context.
  2. Targeted Search: For continuing conversations that require context, the system performs a targeted search for the most relevant memories, optionally using a deep query for more complex questions.
  3. Acknowledge Only: When a client specifies that no context is needed, the system simply acknowledges the message and processes it in the background, optimizing for speed.

​
Opinionated Context Flows

While our primary jean_memory tool provides a balanced approach, the underlying tools can be composed into highly specialized, opinionated flows to solve specific problems. Below are a few examples of what’s possible.
This flow is designed for an AI assistant that needs to provide a user with a summary of relevant information before they even ask. It’s perfect for a morning briefing or preparing for a meeting.
This flow is for tasks that require a comprehensive understanding of a large corpus of information, like a collection of research papers or project documents.
This flow is for an AI that learns a new skill or topic in real-time based on user interaction, getting progressively smarter with each turn of the conversation.

​
Create Your Own Flow

The true power of Jean Memory is its flexibility. The primitive tools (store_document, search_memory, add_memories, etc.) are the building blocks for you to create your own context engineering flows tailored to your specific use case. Whether you’re building a hyper-personalized tutor, a strategic research agent, or something entirely new, our toolset provides the foundation. Want to build a custom flow? Reach out to our team, and we’ll be happy to show you how to get started.
⌘I