GitHub Repo

GitHub 熱門專案、開源工具、框架、runtime 與重要 release。

LangChain 1.4 Brings MCP into the Core Package with Support for Stateless Protocols, Tool List Caching, and Mid-Execution Elicitation

The new `langchain.mcp` uses FastMCP to unify connections, authentication, and protocol negotiation, replacing the standalone adapter package. Servers can request additional information or approval during tool execution, while LangChain maps the pause point to a resumable LangGraph interrupt.

2026-09-06

AI 安全

GitSpawn Exploits Repository-Local Git Configuration to Bypass Execution Approvals in Seven Coding Agents

Malicious projects can place commands in `.git/config` under `core.fsmonitor`, then wait for an agent to run `git status` or `git diff` in the background to execute them on the host. Some products have been patched, but the incident shows that tool approvals can still be bypassed by underlying programs if an agent’s sandbox does not cover startup and version-control processes.

2026-09-06

推論系統

SGLang Adds GPU-Free Simulator to Official Documentation for Replaying Inference Traffic with Latency Predictions

SGLang Simulator intercepts runtime components on the CPU to simulate scheduling, batching, and multi-tier KV caches without actually loading models or executing GPU kernels. The team claims errors below 5% on key metrics for two Qwen3 models running on H20 GPUs, but hardware and model coverage remain limited.

2026-09-06

GitHub Repo

Paddock Open-Sources Rust/CUDA Inference Engine Targeting Concurrent AI Agents on a Single NVIDIA GPU

Truespar has released Paddock’s scheduler, paged KV cache, memory management, and custom CUDA kernels, along with OpenAI- and Anthropic-compatible interfaces. The authors’ benchmarks show it outperforming vLLM and SGLang on some single-GPU workloads, but the results have not been independently reproduced, and Paddock does not support tensor parallelism, ROCm, or Metal.

2026-09-06

代理工具

DRACO Assigns Reinforcement Learning Credit to Each Agent Step Using Dynamic Rubrics—Without Accessing Task Answers During Training

A team from IBM Research and Carnegie Mellon University enables a judge model to generate rubrics for each set of trajectories, then redistributes the overall reward to the relevant steps. The method improves performance on AppWorld by 15.9 percentage points, although its correctness still depends on model judgments that have not been manually validated.

2026-09-06

GPT-6 Astra API Adds Asynchronous Tool Calling and Mid-Turn Steering as Cybersecurity Capabilities Reach Critical for the First Time

OpenAI’s new flagship model can continue reasoning while tools run in the background and receive mid-execution corrections over WebSocket. Its stronger agentic capabilities come with new migration constraints—and the first Critical rating for cybersecurity capabilities.

2026-09-06

代理工具

Pydantic AI 2.40 Adds Playback-Calibrated Barge-In Interruptions for Realtime Voice Agents

The new release lets applications interrupt responses when users speak over them based on the number of audio bytes actually played, while keeping conversation state in sync. It also adds an out-of-band prompt queue and an event-listening interface, so realtime agents no longer need to disguise every control signal as a user turn.

2026-09-06

推論系統

llama.cpp 0.4.0 Reduces Large-Model Loading Spikes with Lazy Tensor Loading and Adds Support for the New Qwen3.8 Architecture

The new release adds `--lazy-mode`, a quantization working-memory limit, and row-slab streaming, eliminating the need to load extremely large tensors into RAM all at once. It also introduces preliminary support for Qwen3.8-Flash-Next and Nemotron-3-Puzzle, though performance optimizations and compatibility with some hardware remain incomplete.

2026-09-06

代理工具

Claude Code 2.1.261 Exposes Skills’ Context Costs and Tightens Data Exfiltration Rules in Auto Mode

The new `/skill-doctor` command lists unused skills and their context footprint, helping developers identify prompt costs paid repeatedly on every turn. The update also expands dangerous deletion detection, fixes context loss during session restoration, and treats public chart URLs containing embedded content as external uploads.

2026-09-06

AI research

One-Shot OPD Captures Most Full-Data Distillation Gains With a Single Query, Shifting the Bottleneck to Teacher-Signal Absorption

Researchers found that repeatedly sampling from a single query while giving the student model token-level teacher supervision can, after 300 steps, cover 71.5% of the state reached by full-data training. Adding 16 semantically diverse queries raises coverage to 98.9%, though the findings remain limited to the specified small models, teachers, and benchmarks.

2026-09-06

AI 研究／形式驗證

Claude Uses Multi-Agent Collaboration to Formalize Fermat’s Last Theorem in Lean, Releases 13-Million-Line Proof

Anthropic says dozens of Claude agents completed the first fully computer-checkable, end-to-end proof of Fermat’s Last Theorem in 11 days. The real breakthrough lies not only in model capability, but in how Prove2Me uses theorem dependency graphs, independent compilation, and search mechanisms to manage long-running collaboration.

2026-09-05

推論系統

NVIDIA PAIR Dispatches Cross-Device Inference Through a Single Local Endpoint, but Does Not Pool GPU Memory

The open-source Personal AI Router can schedule independent Ollama or LM Studio requests across Windows, Linux, and macOS nodes. It is suited to increasing multi-agent concurrency, but it is not a distributed model execution engine: a single model must still fit entirely on one machine.

2026-09-05