Decoding the Next Frequency
of Artificial Intelligence.

OpenAI models and Codex are now accessible through Oracle Cloud using existing cloud commitments, enabling enterprise-grade AI deployment with security and governance.

This facilitates streamlined AI adoption for enterprises already invested in Oracle Cloud by simplifying access to OpenAI’s capabilities while ensuring compliance with enterprise security and governance requirements.

Maestro is a lightweight, research-grounded multi-agent orchestrator designed to coordinate AI coding agents across multiple LLM platforms like Claude Code, Gemini, Codex, and Cursor, with zero dependencies.

This tool facilitates integrating and orchestrating multiple AI coding agents from different providers seamlessly, enhancing AI-assisted software development workflows and experimentation with multi-agent systems.

Agent Service Manifest (ASM) is a JSON schema specification to define and price AI agent services for better service discovery and interaction.

It enables interoperable service discovery and fair pricing mechanisms for AI agents, fostering an agent economy and efficient multi-agent collaboration.

Ariadne is a Python-based AI agent tool that enables LLMs to provide precise answers using pre-processed code knowledge across multiple repositories and languages without needing to rescan files each session.

This approach reduces redundant computation and context rebuilding for LLMs, improving response speed and accuracy when handling large, polyglot codebases, which is critical for scalable AI code assistance.

This GitHub repository contains experimental code showcasing AI-powered developer relations tools using agent frameworks and LLMs.

It explores practical AI applications in improving developer engagement and support, a niche use case showing real AI integration with developer tools.

moltstream is a tool that deploys autonomous AI streamers on Kick, automating chat, voice, avatar, and OBS streaming setup quickly using AI agents.

This project demonstrates practical deployment of autonomous AI agents in real-time interactive streaming, showcasing AI's expanding capability in content creation and live user engagement automation.

The paper introduces Context-Driven Incremental Compression (C-DIC), a method for efficient and robust multi-turn dialogue modeling that maintains scalable context memory across many dialogue turns without losing fidelity.

This approach addresses key challenges in scaling dialogue agents to long conversations by reducing redundant computation and memory usage while preserving dialogue coherence and quality over time, enabling more practical and capable conversational AI systems.

DIRECT is a routing framework that smartly allocates test-time compute for embodied agents, optimizing success rate and latency more efficiently than naive scaling. It improves embodied planning in robotics by dynamically selecting compute resources based on scene context.

Naively increasing test-time compute in embodied agents is costly and yields diminishing returns; DIRECT enables more efficient, cost-effective deployments by tailoring compute usage to task context, pushing frontier performance closer to real-world robotic applications.

Doc-to-Atom introduces a compositional memory mechanism that decomposes documents into reusable LoRA adapters for efficient long-context reasoning in LLMs.

This approach addresses key limitations in handling long documents by enabling modular, query-specific adapter composition, which enhances scalability, reduces interference, and improves multi-step reasoning efficiency in large language models.

This paper proposes a new design for Mixture-of-Experts (MoE) routers using Manifold Power Iteration to align router rows with principal singular directions of expert matrices, improving MoE model effectiveness.

By theoretically and empirically improving router design, this approach enhances the token-to-expert affinity calculation in MoE models, potentially resulting in more efficient routing and better model capacity utilization, which is critical for scaling large models.