2026年7月29日 技术热点总结

项目简介：Open Frontier Intelligence

核心介绍：📰 Tech Blog | 📄 Full Report Kimi K3 is an open-weight, native multimodal agentic model and our most capable model to date. It is a 2.8T-parameter model built on Kimi Delta Attention (KDA) and Attention Residuals (AttnRes), with native vision capabilities and a 1-million-token context window. It is the world’s first open 3T

项目数据：⭐ 3,366 Stars，🍴 277 Forks

项目简介：TypeScript-to-Native Compiler

技术栈：TypeScript

核心介绍：function fib(n: number): number { return n < 2 ? n : fib(n – 1) + fib(n – 2); console.log(fib(30)); $ scriptc build fib.ts && ls -la fib No changes to your code. No annotations, no dialect — the same TypeScript you run on Node, type-checked by the real TypeScript compiler and compiled to native. What compiles behaves byte-for-byte like Node.

项目数据：⭐ 2,111 Stars，🍴 40 Forks

项目简介：AgentENV (AENV) is a distributed platform for running agent environments at scale.

技术栈：Rust

核心介绍：Running agent environments at scale 📖 Full documentation AgentENV (AENV) is a platform for running agent environments at scale, powering agentic RL training for Kimi K3.

项目数据：⭐ 1,426 Stars，🍴 127 Forks

项目简介：Permanent memory for AI agents. A 426-token prompt, a script, plug and play.

技术栈：Python

核心介绍：Permanent memory for AI agents. A 426-token prompt, a script, plug and play.

项目数据：⭐ 797 Stars，🍴 48 Forks

项目简介：MoonEP: A Perfectly Balanced Expert Parallelism Library via Dynamic Redundant Experts

技术栈：Python

核心介绍：MoonEP is an Expert Parallelism communication library that keeps token loads perfectly balanced across ranks via dynamic redundant experts. 1. Perfect balance: every rank receives exactly S × K tokens, no matter how skewed the routing is. A small number of redundant experts is planned online from the current router outputs and prefetched before expert computation; their gradients are reduced back to their home ranks in the backward pass.

项目数据：⭐ 781 Stars，🍴 77 Forks

项目简介：An archive of public Claude and Grok conversations, exported from their share links as plain markdown, plus the two scripts that produce it.

技术栈：Python

核心介绍：An archive of public Claude, Grok, Kimi, Qwen, and DeepSeek conversations, exported from their share links as plain markdown, plus the five scripts that produce it. A handful of other conversations, added by hand from sources with no public share link, live alongside them. 1,528 conversations, 16,394 messages, ~66 MB. scripts/claude_share_export.py exporter for claude.ai shares scripts/grok_share_export…

项目简介：runs anywhere. uses anything

技术栈：TypeScript、agentic-ai、ai、ai-agent、ai-coding、ai-coding-agent、ai-coding-agents、ai-coding-assistant、anthropic

核心介绍：An open-source coding agent for the CLI. Cloud APIs, gateways, and local models — same tools, same agents, same workflow.

项目数据：⭐ 581 Stars，🍴 86 Forks

项目简介：Local-first Agent Work Intelligence for coding agents: usage truth, recorded work, and honest joins. Read-only over coding-agent logs; zero-JavaScript localhost dashboard.

技术栈：Python

核心介绍：agentacct is local-first Agent Work Intelligence for coding agents. It reads the session logs that Claude Code and Codex already write on your machine, joins them with the work each session records as it goes, and shows the result — tokens, estimated cost, tasks, and evidence — on a local dashboard. Screenshots show a synthetic demo workspace; your dashboard renders your machine’s real l…

项目简介：Ultraminimalist macOS recording + transcription.

技术栈：Swift

核心介绍：A minimal, fully local macOS meeting recorder + transcriber. One menu-bar click records your mic and all system audio as two separate tracks; when you stop, quill transcribes both on-device and writes a speaker-tagged transcript. Nothing ever leaves the machine. Named for the feather. Sibling of parrot, same skeleton: single Swift binary, menu-bar tray, no app bundle.

项目数据：⭐ 517 Stars，🍴 26 Forks

项目简介：Free, open-source macOS alternative to BetterDisplay and Lunar: a lightweight menu bar app with sharp HiDPI/Retina scaling for external monitors (no more blurry or tiny text), plus brightness (DDC), virtual displays, presets, and color.

技术栈：Swift、4k、apple-silicon、brightness、ddc、display、display-manager、hidpi、mac

核心介绍：Crisp is a free, open-source alternative to BetterDisplay and Lunar: a lightweight, native macOS menu bar app for managing external displays. It’s built to feel like an expanded version of the Mac’s built-in display menu: the controls you already use, plus the…

Bitcoin price prediction on sub-daily timescales is a hard open problem in computational finance. Bitcoin exhibits fat-tailed returns, non-stationary dynamics, and a price discovery process influenced by social discourse on Reddit and Twitter. Conventional approaches fuse OHLCV technical features with sentiment via static concatenation, applying identical fusion weights regardless of market state. This is inconsistent with the behavioural finance literature, which shows that retail sentiment is most predictive during volatile periods and noisy during calm ones. This paper proposes Regime-Aw…

Recent advances in powerful text-to-image generation models have made it increasingly important to develop test-time methods that modify the sampling trajectory to produce images more faithful to complex compositional prompts. We present TILT, a training-free framework for compositional text-to-image generation via test-time reward alignment. We interpret compositional failures as overlap modes between joint and single-concept distributions, and define a reward that favors samples where all concepts are jointly present. This reward is intrinsic to the base model and does not require any ext…

Large Vision-Language Models (LVLMs) remain bottlenecked by massive computational footprints, precluding their deployment on resource-constrained edge devices. While efforts to compress LVLMs focus heavily on vision token reduction or smaller language models, the vision encoder is largely overlooked, typically deployed as a monolithic, computationally heavy feature extractor. Moreover, there is no previous effort that designs a vision encoder for LVLMs directly optimized for on-device latency. In this paper, we present UltraViT, a vision encoder for LVLMs, explicitly designed and optimized …

Many recent robot policies pursue stronger control by using large pretrained vision-language models (VLMs) as the action backbone. We introduce WorldDiT, a unified diffusion transformer architecture that couples action generation with visual world modeling and achieves strong performance without a large pretrained VLM action backbone. During training, a single diffusion transformer generates continuous action chunks and predicts normalized RGB patch targets from future camera frames. Across four LIBERO simulation suites, WorldDiT lies on the reported Pareto frontier for total model paramete…

We introduce a vocabulary for automated research systems built from one or more agents to make their design choices easier to describe and compare. The vocabulary specifies 1) who the agents are, 2) what operations are available in the system, 3) who may invoke them, 4) how agents communicate, 5) what information is visible within and across runs, 6) how the next action is chosen, 7) how a run begins, and 8) how outputs are evaluated. A trajectory records one run from the input task to the returned artifact. Because agents, operations, and initialization may be stochastic, repeated runs on …

Parametric retrieval enables LLMs to retrieve tools implicitly by assigning each API a unique virtual token and training the model to generate it via constrained beam search. Toolsense shows that this regime has two critical drawbacks: it destroys parametric tool knowledge during training, and its beam-search decoding is too slow for real-time deployment. We introduce TRACE (Tool Retrieval via Augmented Chain-of-thought and Enterprise rules), a two-stage curriculum that resolves this dissociation. Stage 1 reuses the multi-format memorization SFT from ToolSense to seed tool knowledge with Lo…