Agentic Memory Unifies Long Term and Short Term Memory for LLM Agents

Here is your today’s AI Dev Brief from Marktechpost, covering core research, models, infrastructure tools, and applied updates for AI developers and researchers.

AgeMem is a new agentic memory framework that integrates long term and short term memory management directly into an LLM agent policy through tool based actions. Instead of using external controllers or fixed heuristics, the agent chooses when to call tools such as ADD, UPDATE, DELETE, RETRIEVE, SUMMARY and FILTER in the same action space as text generation. The model is trained with step wise Group Relative Policy Optimization in a three stage setup that first builds long term memory, then learns short term context control under distractors, and finally performs integrated reasoning for the target task. A unified reward combines task accuracy, context quality and memory quality. On ALFWorld, SciWorld, BabyAI, PDDL tasks and HotpotQA, AgeMem on Qwen2.5-7B and Qwen3-4B improves success rates, memory quality and token efficiency over existing memory baselines........ Read the full analysis/article here.

AI teams spend millions on proprietary RL environments for training terminal agents, while CAMEL AI community group just open sourced SETA, Scaling Environments for Terminal Agents, as a full training stack. SETA ships 400 containerized terminal training tasks with instruction, Dockerfile, and verifiable tests, a state of the art CAMEL terminal agent harness on Terminal Bench 1.0 and 2.0 using Claude Sonnet 4.5 and GPT 4.1 within their model families, and the complete RL pipeline plus released SETA RL Qwen3-8B model weights. Together this gives devs a reproducible way to train, debug, and benchmark Unix style terminal agents using open infrastructure instead of closed commercial environments....... Read the full analysis/article here.

▶ [Open Source] Rogue: An Open-Source AI Agent Evaluator worth trying Codes & Examples

▶ A Coding Guide to Demonstrate Targeted Data Poisoning Attacks in Deep Learning by Label Flipping on CIFAR-10 with PyTorch Codes Tutorial

▶ How to Build a Proactive Pre-Emptive Churn Prevention Agent with Intelligent Observation and Strategy Formation Codes Tutorial

▶ A Coding Guide to Design a Complete Agentic Workflow in Gemini for Automated Medical Evidence Gathering and Prior Authorization Submission Codes Tutorial

100+ more open codes/notebooks here ➡️

180+ AI releases— 3 months of hard work for open source community.
→ [Check it out]

Partner with us 2026 Report/Platform: [email protected]