Seven Open Source Solutions to Enhance AI Agents' Memory

Memory is becoming a key building block for agents capable of chaining tasks and retrieving their context. Seven open-source projects offer different pathways, from temporal graphs to state persistence, including reflection and portability between coding tools. Here’s an overview of the available options and their use cases.

Defining Context and Memory for Sustainable Agents

Context refers to what an agent can consult at a given moment, while memory denotes what it retains for later. With AI memory, agents can remember important elements from previous conversations and use them in future interactions. An agent has the ability to record facts, preferences, actions taken, or relevant information, and then reuse them to generate more tailored responses that take context into account. When agents handle longer and more complex missions, their ability to memorize, retrieve, and update past experiences evolves a stateless model into a system that capitalizes on what has already been learned. By storing and reusing relevant information, they ensure continuity, leverage acquired knowledge, and can improve their efficiency during repeated exchanges.

Knowledge Graphs and Time: Cognee and Graphiti

Several projects rely on graphs to structure a rich memory. Cognee transforms documents, code, and conversations into a connected and queryable memory. Its pipeline combines vector search and graph relationships, allowing agents to retrieve information by meaning and by connections between concepts. This approach explicitly preserves the links between pieces of information. The repository is available on GitHub at https://github.com/topoteretes/cognee. Graphiti applies temporal knowledge graphs. Information and relationships are modeled as evolving, which helps assistants maintain a changing context, distinguish old from new, and keep an accurate history of changes. Its code is accessible at https://github.com/getzep/graphiti. These approaches, which fit within workflows ranging from compact facts to elaborate graphs, address the same fundamental issue of agent memory.

State Persistence and Continuity: Mem0 and OpenViking

Other solutions target the preservation of state over time. Mem0 presents itself as a general-purpose memory layer: it stores and retrieves memories of users or agents to maintain information between sessions, including facts, preferences, and interactions, without relying solely on the current conversational context. Its GitHub repository is available at https://github.com/mem0ai/mem0. OpenViking emphasizes state persistence and recovery over interactions. Designed to organize context, it allows an agent to reuse previous information, maintain continuity between sessions, and retrieve relevant elements when a similar task arises or to build on previous work. The project is accessible at https://github.com/volcengine/OpenViking. Both projects address the same need for operational continuity of agents.

Proactivity and Reflection: Hindsight and memU

Hindsight articulates long-term memory around three complementary functions: memory, recall, and reflection. It aims for agents to retain information and leverage past experiences for future decisions, building a persistent understanding of exchanges beyond the isolation of each session. Its code is available at https://github.com/vectorize-io/hindsight. For its part, memU views accumulated experience as a structured set of knowledge that can be retrieved. Its goal is to provide memory that is both durable and anticipatory, rather than a simple mechanism for occasional recall, enabling the automatic emergence of useful knowledge whenever a new task requires it. The source code can be found at https://github.com/NevaMind-AI/memU. Both projects approach memory as an active resource, not just as a history to consult.

Continuity Between Coding Environments: OpenMemory

OpenMemory targets context portability in software development. The tool imports and exports coding sessions between agent environments such as Claude Code, Codex, and OpenCode, to avoid losing context during tool changes. This portability allows for resuming work without reconstructing the state of the conversation or task. The repository is accessible at https://github.com/mem0ai/openmemory.

Use Cases Beyond Chatbots

Agent-side memory is not limited to conversational interfaces. It is used in programming-specialized agents, research tools and platforms, as well as versatile assistants. In these situations, memory allows for the storage of data related to missions, exchanges, discoveries, or user preferences, and then retrieves them when they become useful again. The common challenge across all identified projects remains the same: to provide agents with a stable foundation of reusable knowledge.