代理工具

Agent、MCP、coding agent、工作流自動化與評測。

OpenClaw 2026.9.3 Rehearses Upgrades in Isolation, but Simultaneously Breaks Compatibility with Node and Plugin Interfaces

The new OpenClaw release validates the core, plugins, and migrations in an ephemeral candidate state before activation, with a separate check serving as the final gate. It also retains records of failures and rollbacks. At the same time, the upgrade drops support for Node 22/25 and several SDK interfaces; one report indicates that an agent may still lose semantic session continuity even after every structural check passes.

2026-09-10

AI 安全

Red Hat AI 3.5 Brings OpenShell into OpenShift to Constrain Agent Side Effects with Kernel-Level Policies

OpenShift AI 3.5 integrates OpenShell as a Developer Preview, using Landlock, seccomp, network namespaces, and L4/L7 policies to restrict agent access. This means file, system-call, and network side effects following prompt injection no longer depend solely on model guardrails, although the default compatibility mode and preview status may still leave gaps.

2026-09-10

GitHub Repo

MiniCPM5-2B Opens Training Data and On-Device Deployment Formats, but Quantized-Version Sampling Defaults Draw Scrutiny

OpenBMB has released a 2B dense model with a 131K-token context window and tool-calling capabilities, along with staged checkpoints, training data, and multiple quantization formats. Third-party evaluations affirm its competitiveness within its class, but community testing suggests the official sampling settings may cause GGUF quantizations to fall into repetitive generation loops.

2026-09-10

代理工具

DeepSeek Harness 0.1.5-alpha.1 Preserves the KV Cache for Dynamic Prompts but Reworks Session and Plugin Interfaces

The new release lets supported models continue using the KV cache when the system prompt is dynamically updated, reducing recomputation costs in long-running sessions. The tradeoff is a set of backward-incompatible changes across Session V3, Agent, and Inbox APIs, requiring plugins and custom log tooling to be revalidated.

2026-09-09

代理工具

Pydantic AI 2.41/2.42 Integrates Codex and Copilot, Tightens Validation of Deferred Tool Approval Data

Two consecutive releases add a direct image generation API and providers for ChatGPT/Codex subscription access and GitHub Copilot’s compatible endpoint. Version 2.42 also rejects malformed deferred tool approval results, so upgrading may cause agent workflows previously accepted under lenient validation to fail outright.

2026-09-09

RAG 與資訊檢索

Q2D-Web Tests Agentic RAG on 190 Million Documents, Finds Supporting Queries a Retrieval Weakness

Perplexity’s Q2D-Web uses nearly 70,000 machine-rewritten queries derived from real conversations to evaluate whether first-stage retrievers can recover evidence at web scale. Experiments show that neural retrievers generally struggle more with supporting queries generated by agents, while reducing the corpus preserves system rankings but inflates absolute scores.

2026-09-09

NeoHorse-1 Feeds Agent Execution Traces Back into Model Training, but “Recursive Self-Improvement” Remains a Single-Round Prototype

NeoHorse-1 uses a router to collect traces from different models completing agentic tasks, then trains 4B and 9B models through curriculum-based SFT and on-policy distillation. Its official aggregate scores exceed those of same-size Qwen3.5 baselines, but it has yet to demonstrate that repeated rounds of self-updating can deliver sustained gains or prevent data drift.

2026-09-09

Applied AI research

EastAsiaClimateExtremes Turns East Asian Heatwaves, Heavy Rainfall, and Marine Heatwaves into Weekly AI Labels

The APEC Climate Center has released an East Asian extreme climate dataset aligned with ECMWF subseasonal reforecasts, covering gridded labels, event intensity, and analysis code. It reduces the preprocessing cost of creating AI prediction targets, but its 1.5-degree grid, biases in reanalysis data, and unclear licensing status limit direct deployment.

2026-09-09

代理工具

GitHub Brings Copilot’s Enterprise Sandbox to JetBrains, With Central Policies to Restrict Credentials and Network Egress

GitHub Copilot’s JetBrains plugin can now apply centrally managed enterprise sandbox policies governing access to files, networks, development tools, and the macOS Keychain. The new feature extends agent governance into the IDE, but it remains in public preview, and agent servers and agent egress are not all covered by the same boundary.

2026-09-09

AI 安全

Meta Muse Uses Isolated Execution Units, a Credential Broker, and Sentinel to Control Every External Side Effect of Its Agents

Meta’s personal agent Muse runs inside a dedicated VM for each user. The main agent cannot see real credentials or obtain network egress on its own. The architecture treats prompt injection as a threat that cannot be fully eliminated, but stronger Confidential VMs and externally verifiable audits have yet to be formally delivered.

2026-09-09

推論系統

Cohere Packs the Entire Decoding Step Into a Single CUDA Megakernel, Boosting End-to-End H100 Throughput by 25% to 41%

Cohere has open-sourced a dedicated inference engine for North Mini Code that replaces per-operator launches and full-GPU synchronization with a persistent megakernel. Official benchmarks show higher end-to-end decoding throughput than vLLM on a single H100, though support is currently limited to specific models, hardware, and small batch sizes.

2026-09-09

代理工具

Agno 3.0.7/3.0.8 Publishes Agent Knowledge as Versioned Pages and Closes a Session Ownership Gap

Agno adds atomically publishable page knowledge with revision-pinned reads, alongside a PublicSurface that limits inputs and outputs for Agent, Team, Workflow, and MCP services. Version 3.0.8, released hours later, adds full-page reads and ownership checks for SQLite batch writes; upgrades must also account for changes to HNSW search semantics.

2026-09-09