Instant LLM Updates with Doc-to-LoRA and Text-to-LoRA

Sakana AI on Twitter / X

We’re excited to introduce Doc-to-LoRA and Text-to-LoRA, two related research exploring how to make LLM customization faster and more accessible.https://t.co/wGKDNhBcJXBy training a Hypernetwork to generate LoRA adapters on the fly, these methods allow models to instantly… pic.twitter.com/gId3J6hgEr— Sakana AI (@SakanaAILabs) February 27, 2026

Memory in Agents: What, Why and How

LLM memory gives language models persistent context across sessions. Learn how it works, how it differs from RAG and context windows, and how to add LLM memory to your agents with Mem0.

MemGPT: Towards LLMs as Operating Systems

Large language models (LLMs) have revolutionized AI, but are constrained by limited context windows, hindering their utility in tasks like extended conversations and document analysis. To enable using context beyond limited context windows, we propose virtual context management, a technique drawing inspiration from hierarchical memory systems in traditional operating systems that provide the appearance of large memory resources through data movement between fast and slow memory. Using this technique, we introduce MemGPT (Memory-GPT), a system that intelligently manages different memory tiers in order to effectively provide extended context within the LLM's limited context window, and utilizes interrupts to manage control flow between itself and the user. We evaluate our OS-inspired design in two domains where the limited context windows of modern LLMs severely handicaps their performance: document analysis, where MemGPT is able to analyze large documents that far exceed the underlying LLM's context window, and multi-session chat, where MemGPT can create conversational agents that remember, reflect, and evolve dynamically through long-term interactions with their users. We release MemGPT code and data for our experiments at https://memgpt.ai.

Pieces | Infinite Artificial Memory for your Digital Workers and Agents

Pieces is your AI companion that captures live context from browsers to IDEs and collaboration tools, manages snippets and supports multiple llms - all while processing data locally for maximum control.

mem-agent: Equipping LLM Agents with Memory Using RL

The insights and the technical report behind Mem-Agent, our 4B model for persistent memory in LLMs

The rebel alliance

This blog is co-authored with Zoe Weinberg and Matt Hawes at ex/ante, and is a follow-up to our first blog post on the topic, 'You don't own your memory.' We need an open architecture that puts us in control of our memories while making their exploitation technically impossible. But how will this shift happen? In order to discover possible implementations, we must understand how our data informs LLMs. The three predominant context engineering techniques are prompt design, retrieval-augmented ...

Sakana AI on Twitter / X

We’re excited to introduce Text-to-LoRA: a Hypernetwork that generates task-specific LLM adapters (LoRAs) based on a text description of the task. Catch our presentation at #ICML2025!Paper: https://t.co/2FRiVF1UXJCode: https://t.co/rx4G7dq1SWBiological systems are capable of… pic.twitter.com/UdUYfqRXBS— Sakana AI (@SakanaAILabs) June 12, 2025

The Bitter Lesson of LLM Extensions

From ChatGPT Plugins to Agent Skills, a look at how we've been trying (and failing) to extend LLMs for the last three years.

Large language model

A large language model (LLM) is a neural network trained on a vast amount of text for natural language processing tasks, especially language generation. LLMs can typically generate, summarize, translate, and analyze text in many contexts, and are a foundational technology behind modern chatbots.[1] Biased or inaccurate training data can make an LLM's output less reliable.[2]

DREAM — Dynamic Retention Episodic Architecture for Memory

Modern AI systems lack persistent, user-specific episodic memory. Existing approaches rely on short-term context windows, shallow preference storage, or static conversation logs that do not scale and cannot preserve meaningful long-term continuity. This paper introduces DREAM (Dynamic Retention Episodic Architecture for Memory), a scalable, opt-in, episodic memory framework designed to work with current LLM and agent architectures. DREAM integrates episodic summarization, user-controlled opt-in memory, semantic retrieval via per-user vector indexes, an adaptive retention mechanism that expands TTL based on user engagement, and horizontal sharding of orchestrators and storage for large-scale deployments. The paper details the architecture, components, data flows, and implementation examples, and argues that DREAM provides a practical path toward AI systems capable of consistent, privacy-aligned long-term reasoning. This project has been extended with a conceptual analysis and simulation of a "DREAM-as-a-Support" (DaaS) hybrid layer. This extension demonstrates DREAM's architectural extensibility, reframing it from a standalone framework into a foundational platform component. The DaaS model provides core memory governance such as adaptive retention (ARM) and user-centric opt-in as an on-demand service to complementary cognitive systems, validating the original four-pillar design through a scalable, internal API. Reference Implementation A reference implementation of the DREAM architecture is available as an open-source Python framework:Official Reference Implementation This implementation is intended for experimentation and architectural validation and does not represent a production-ready system DREAM Architecture — Official GitHub Repository

Supporting Our AI Overlords: Redesigning Data Systems to be Agent-First

Large Language Model (LLM) agents, acting on their users' behalf to manipulate and analyze data, are likely to become the dominant workload for data systems in the future. When working with data,...

MemOS: An Operating System for Memory-Augmented Generation (MAG) in Large Language Models

Large Language Models (LLMs) have emerged as foundational infrastructure in the pursuit of Artificial General Intelligence (AGI). Despite their remarkable capabilities in language perception and generation, current LLMs fundamentally lack a unified and structured architecture for handling memory. They primarily rely on parametric memory (knowledge encoded in model weights) and ephemeral activation memory (context-limited runtime states). While emerging methods like Retrieval-Augmented Generation (RAG) incorporate plaintext memory, they lack lifecycle management and multi-modal integration, limiting their capacity for long-term knowledge evolution. To address this, we introduce MemOS, a memory operating system designed for LLMs that, for the first time, elevates memory to a first-class operational resource. It builds unified mechanisms for representation, organization, and governance across three core memory types: parametric, activation, and plaintext. At its core is the MemCube, a standardized memory abstraction that enables tracking, fusion, and migration of heterogeneous memory, while offering structured, traceable access across tasks and contexts. MemOS establishes a memory-centric execution framework with strong controllability, adaptability, and evolvability. It fills a critical gap in current LLM infrastructure and lays the groundwork for continual adaptation, personalized intelligence, and cross-platform coordination in next-generation intelligent systems.

Solving a Million-Step LLM Task with Zero Errors

LLMs have achieved remarkable breakthroughs in reasoning, insights, and tool use, but chaining these abilities into extended processes at the scale of those routinely executed by humans,...

LoRA Learns Less and Forgets Less

Low-Rank Adaptation (LoRA) is a widely-used parameter-efficient finetuning method for large language models. LoRA saves memory by training only low rank perturbations to selected weight matrices. In this work, we compare the performance of LoRA and full finetuning on two target domains, programming and mathematics. We consider both the instruction finetuning (approximately 100K prompt-response pairs) and continued pretraining (20B unstructured tokens) data regimes. Our results show that, in the standard low-rank settings, LoRA substantially underperforms full finetuning. Nevertheless, LoRA better maintains the base model's performance on tasks outside the target domain. We show that LoRA mitigates forgetting more than common regularization techniques such as weight decay and dropout; it also helps maintain more diverse generations. Finally, we show that full finetuning learns perturbations with a rank that is 10-100X greater than typical LoRA configurations, possibly explaining some of the reported gaps. We conclude by proposing best practices for finetuning with LoRA.