Tag
Rohan Arun built ARC (Agent Relay Chat), a lightweight HTTP-based relay that lets AI agents like Codex and other bots join shared chat rooms to collaborate on tasks, with invite codes and agent-name registration for real-time coordination.
SERA（Self-Evaluating Recursive Agents）提出一个联合训练单一策略以实现子任务分解、执行与评估的框架，通过可验证结果对齐评估准则并引入叶覆盖奖励，显著优于递归智能体基线，表明学会评估能提升求解能力。
这篇论文提出了一种基于百分位控制的高速公路场景生成方法，通过条件扩散模型与采样时风险引导，将 corner case 的极端程度量化为其在条件风险分布中的百分位，从而实现对生成场景风险程度的精细调控。
Meta 论文提出 RankEvolve，通过可执行操作协议（EOP）将研究阶段编译为受运行时强制执行的状态机，并让 Claude Code 与 Codex 作为节点互相审查修复，在预算持平下将执行准确率从 45.8% 提升到 62.5%，且在 MovieLens-20M 上改进推荐模型的 NDCG@10 达 4.48%。
A user reports combining four AI models — Claude Opus 5.5, Sonnet 5.5, Jev, and Dots — to build a fully operational 24/7 automated trading bot in about 20 minutes, with strategy generation, code implementation, risk gating, and runtime monitoring divided across the models.
The author dissects Claude Code's source code to explain how its multi-agent system layers the Agent tool, a task registry with a command queue, disk-based mailbox teammates, and a deterministic JS workflow engine — all funneled through a shared query executor with isolated contexts and message-only communication.
NoHope has launched 'swarm,' an experimental web-based project focused on growing and controlling swarms, currently accessible via nohope.io.
An hour-long conversation between Poteto (creator of pstack, SpaceX) and Matt Pocock revealing how they merge 2,500 PRs per month: from distilling repeated AI corrections into Skills, to giving Agents the ability to verify their own work, using tools and lint constraints to cut down repetitive effort, all the way to multi-Agent inner and outer loops plus spot-check reviews.
Zineng Tang promotes the LSEI workshop at COLM 2026 on Oct 9, exploring how language models can learn by interacting with the world and other agents, featuring a speaker lineup and open discussion.
MegaCapybara 是一个针对 NVIDIA RTX 5090 优化的开源 LLM 推理引擎，支持 Qwen3.8-27B，单智能体可达 500+ tokens/s，多智能体并发最高 2000+ tokens/s，具备投机解码、共享上下文池和智能 VRAM/RAM/磁盘缓存管理。
The authors introduce Choir, an open, modular protocol for distributed multi-agent autoformalization that decomposes projects into GitHub-based tasks with deterministic trust-gate verification, supporting Lean 4, Isabelle, and Rocq; it includes a preprint and an open-source repo funded by DARPA's expMath program.
Google Research presents Cogentic, a multi-agent harness built on Gemini that runs an iterative prove–verify loop to discover proofs for open problems in theoretical computer science, producing expert-verified novel results on five open problems in online learning, auction theory, and mechanism design.
A new Google paper introduces Cogentic, a multi-agent orchestration system that enables Gemini to discover new proofs for 5 previously unsolved math problems by combining parallel idea generation, strict step-level verification, and persistent shared notes.
A field report on an open-source experiment testing how multiple AI coding agents share a codebase: isolated branches produced semantically conflicting but textually mergeable work in every run, while a shared working directory and a lightweight 'decision model' referee (Jev) improved coordination, and unsupervised agent-to-agent messaging emerged on its own.
The paper introduces Hermes, a family of configurable harnesses that let models control context allocation across inference-time compute, and Hermes-Learn, a two-stage framework that trains models to make these contextual reasoning decisions adaptively, closing the gap for smaller open-source models.
The author compares AgentWS workspace infrastructure against protected Git worktrees for multi-agent file updates, showing identical outcomes but far less custom code with AgentWS. The post highlights ACLs, recovery after worker failures, and conflict handling as requirements beyond plain Git.
The paper introduces Collective Bias Mitigation (CBM), a framework that reduces LLM bias by fine-tuned model behavior and knowledge sharing among diverse LLMs, using Debating and Committee routing topologies to balance bias reduction and inference cost.
CodeAF is a coding harness designed for open-source models, aiming to deliver frontier-grade coding performance at a fraction of the cost while offering a unified window to manage and monitor multiple coding agents across projects.
A Meta and University of Washington paper introduces Context Language Models (CLMs), where an agent treats its own context as an editable file it manages via Bash. Zero-shot and RL-trained variants show better accuracy with fewer FLOPs on BrowseComp-Plus and multi-agent tasks, plus a Suffix Cache Reuse trick to offset cache invalidation costs.
Multiple teams independently built AI agents that each maintained isolated memory over the same accounts, causing conflicting actions at renewal time. The fix was moving account history to a shared live store—DevRev's Computer synced with HubSpot, Gmail, and Slack—accessed by all agents via MCP.