All published AI Radar articles.

Moltis is a secure, persistent personal agent server built in Rust. It offers sandboxed execution, supports multiple LLM providers, and integrates with various communication platforms like Telegram, WhatsApp, and Discord. Designed for self-hosting on user hardware, it emphasizes security and ease of deployment with a single binary.

SylphxAI has released PDF Reader MCP, a local-first, native Rust tool designed to give AI agents "eyes" for PDF documents. It extracts structured text, tables, and OCR data, providing visual evidence and page-level citations.

A new Rust project, mcp_agent_mail_rust, offers a server for multi-agent coordination. It features 34 tools, a Git-backed archive, SQLite indexing, and an interactive TUI console.

Crabbuild Compass is a native, local-first knowledge graph engine for code and project artifacts, built in Rust. It aims to evolve beyond its inspiration, Graphify, offering advanced capabilities for developers.

Crabbuild has released Compass, a native, local-first knowledge graph engine designed for code and project artifacts. Inspired by Graphify and built with Rust, it aims to evolve beyond its predecessor.

Researchers have introduced Diagram-MMU, a new benchmark designed to evaluate the capabilities of Multimodal Large Language Models (MLLMs) in understanding and processing scientific diagrams. The benchmark includes a dataset of diagrams and associated questions, focusing on tasks like diagram-to-code generation and question answering.

OpenAI's research indicates that enterprises are increasingly adopting agentic AI. The study highlights the use of tools like ChatGPT and Codex in this adoption, with leading companies showing faster progress.

OpenAI has partnered with AWS to offer its Daybreak cybersecurity models through Amazon Bedrock. This integration aims to enhance enterprise security workflows by providing specialized AI capabilities.

Flyto2 Core is an open-source execution kernel designed for automation and AI-agent workflows. It offers a registry of modules, MCP-native transport, and features like YAML recipes, evidence capture, and versioning.

Researchers have introduced an 'Agentic Self-Improvement' framework to enhance control and reliability in Image-to-Video (I2V) generation. This approach reframes video synthesis as a goal-directed optimization process, addressing the trial-and-error nature of current black-box models.