ProjectMem: Local-First Memory System for AI Coding Agents
Researchers have developed ProjectMem, an open-source memory layer that helps AI coding assistants retain context across sessions by logging development events and preventing repeated mistakes. Current AI coding agents waste significant tokens (5,000-20,000 per session) re-reading files and re-deriving decisions because they lack persistent project memory. The system addresses a practical bottleneck in AI-assisted development by combining memory storage with governance features that warn agents before repeating failed fixes.
ProjectMem is a local-first, event-sourced memory system designed to improve the efficiency and reliability of AI coding agents. The system records development activities as an append-only, plain-text event log containing issues, attempts, fixes, decisions, and notes, then deterministically projects this log into compact summaries accessible via the Model Context Protocol (MCP). Beyond simple storage, ProjectMem implements what the authors call "Memory-as-Governance"—a deterministic pre-action gate that alerts agents before they repeat previously failed fixes or modify known-fragile files. The system operates entirely offline with no telemetry and creates an immutable provenance trail for auditable, reproducible AI-assisted development. Implemented as a lightweight Python package with 14 MCP tools and 19 CLI commands, ProjectMem was evaluated through a two-month self-study across 10 projects with 207 logged events.
What's missing
The paper does not provide quantitative performance metrics comparing ProjectMem-augmented agents against baseline agents without memory, such as task completion rates, time-to-completion, or token efficiency gains. The evaluation is limited to a self-study across the authors' own projects, and generalization to diverse development workflows and team settings remains unexplored.
What different sources said
- arXiv cs.AICenter
PROJECTMEM: A Local-First, Event-Sourced Memory and Judgment Layer for AI Coding Agents
Related
New AI Model Improves Prediction of Therapeutic Peptide Function from Protein Sequences
Researchers developed a lightweight CNN classifier that predicts whether peptide sequences have therapeutic properties, trained on a database of 54,655 peptides across 48 functional categories. The model uses a novel negative sampling strategy to reduce false positive rates from over 60% in previous approaches to 2.1%. This advancement could accelerate drug discovery by enabling faster computational screening of peptide candidates before expensive experimental testing.
Study Shows Different Metabolic Stress Models Produce Distinct Effects on Human Neuronal Networks
Researchers tested three common in vitro metabolic stress models on human-derived neuronal networks and found each produced different patterns of neuronal activity and cell damage. The models tested were hypoxia alone, oxygen-glucose deprivation (OGD), and hypoxia combined with glutamate exposure. The findings suggest that choice of experimental model significantly affects results and that combining electrophysiological and structural analyses is important for accurately assessing metabolic stress in stroke research.
New Semi-Supervised Method Improves Single-Cell RNA Sequencing Integration Using Virtual Adversarial Training
Researchers introduced scCRAFT+, a semi-supervised integration method for single-cell RNA sequencing that uses Virtual Adversarial Training to incorporate marker gene information. The method addresses a key limitation of existing approaches: over-mixing of closely related cell subtypes during data integration. The advancement could improve cell type identification and biological interpretation in genomics research.