agent

Agent Reach is a free, open-source tool that gives Claude the ability to read live internet content from 17 platforms (YouTube, GitHub, Reddit, LinkedIn, X, Bilibili, Weibo, etc.) without paid API keys, bypassing official APIs to avoid fees.

Introduces Continuous Interaction Diffusion (CID), a diffusion-native model–runtime architecture that integrates tool interaction into iterative denoising, enabling asynchronous external reads while generation continues without waiting for discrete tool calls.

xAI is launching Grok Bot, AI teammates that can work autonomously across apps and tools in the cloud, available in beta for select subscribers.

Meta releases Muse Glimmer, a 30B open-weight multimodal model optimized for local agent workflows, with permissive Apache 2.0 licensing, 4-bit quantization support, speculative decoding, and broad ecosystem integrations.

Meta introduces Muse Glimmer, a permissively licensed 30B-parameter model optimized for local agent workflows, coding, and tool use, with weights released on Hugging Face.

The author shares an agent-design-review Skill for systematically diagnosing and optimizing Agent architectures, covering Prompt, tool permissions, context, security, memory, evaluation, cost, observability, and more, and outputting evidence-based P0/P1/P2 issues to help avoid common pitfalls.

This paper introduces Macaron-V1, an open continual learning agent-model family using Mixture-of-LoRA to compose specialist adapters on frozen base models, with recursive self-improvement and model-harness co-design.

Author recounts using OpenAI's Codex to migrate data between Snowflake trial accounts, posing the question of whether Codex qualifies as an agent.

A tweet by @tobi demonstrating how to use herdr, a CLI tool, by launching an agent and running `herdr --skill` to create a sci-fi-style hacker terminal setup.

A developer reflects on six months of using AI for code review, finding that vague prompts produce plausible but useless feedback. The fix is treating review as a gated pipeline with explicit context, scoped passes, validation checklists, and adversarial self-critique.

The paper introduces Crafter, an agent for corrective feature discovery that mines the residual of frozen black-box forecasters using compositional search and LLM-generated features, achieving up to 27% error reduction across six datasets and backbones.

RA-CAD presents a state-aware agent for text-to-CAD generation that uses a Generate–Execute–Critique–Rewrite loop, with feedback-driven agent optimization via Group Relative Policy Optimization. It achieves state-of-the-art execution validity and geometric quality on CADFusion and Text2CAD benchmarks.

ReASearch is a unified framework where a single tool-using agent internalizes the search policy for optimizing prompts, programs, and ML workflows, outperforming specialized baselines across 14 tasks.

Introduces The Architect, a Claude Code plugin that generates a complete project blueprint through interviews before writing code, supports 14 project types, and helps avoid architecture going off track and rewriting from scratch.

CSDN founder Jiang Tao will live-stream a discussion on the challenges of taking Agents from demo to truly deliverable, breaking down with InfiniSynapse and InsCode how to connect to real data, complete long tasks, and handle failures.

MemArena is a new ego-centric benchmark for evaluating on-device personal memory assistants, using a MASim agent simulator to generate multi-session conversational worlds and ground truth across recall, reasoning, and trustworthiness dimensions. Initial results show memory-backend choice often matters more than reader scale, and permission-aware access remains a universal challenge.

UrbanAgent is a tool-augmented agent framework that uses LLMs with code execution, API calls, and MCP to handle cross-system urban requests. The authors also introduce UrbanEval, a benchmark for evaluating task results and execution quality, achieving 71% success rate over baselines.

After leaving Shanda Innovation Institute, Lin Kaisen founded a new company focused on long-term memory agents, and mentioned his experience working under the guidance of Boss Chen.

Introduces RubricReviewer, a rubric-driven framework for LLM-based peer review that explicitly generates paper-adaptive rubrics and combines a training-free evidence-gathering agent (Scout) with a trained human-aligned model (Aligner) to produce more comprehensive, discriminative, and robust reviews.

Video-DeepResearch (Video-DR) extends multimodal agents from static images to continuous video streams, introducing a decoupled perception-exploration pipeline and a new benchmark Video-DR-Bench. Their Video-DeepResearch-35B-A3B model achieves 64.0% accuracy, surpassing Claude-4.5-Sonnet, GPT-5, and Gemini 2.5 Pro.