By olud.ai editorial · built from our own tracking data, published daily
The ecosystem gained 270,384 GitHub stars across 10,000+ tracked open-source AI projects today. 65 new projects entered the index. This continues a busy week with 1,931 new projects added and 100 new Hugging Face spaces.
Llama.cpp released build b10369 with support for Pocket-TTS via the mtmd module. Several other projects issued minor bug-fix releases. Price moves affected three major models. Two new projects appeared in the tracking index, one focused on A-share trading agents and one on LinkedIn posting skills.
A community research paper proposing ComBodied Agents received 75 upvotes on Hugging Face. The paper argues that combining software agents with embodied agents creates a more complete human-centric AI system.
Releases of the day
llama.cpp b10369 adds Pocket-TTS support through a transposed convolution approach, enabling text-to-speech within the mtmd inference framework. n8n 2.34.5 fixes TLS options per hop when requests go through a proxy. CrewAI 1.15.15 adds reporting for flow outcome, duration, and human-in-the-loop signals, and fixes a boundary hook event issue. Pydantic AI v1.107.4 re-releases after a build-tooling problem with hatchling 1.32.0 broke the earlier v1.107.3 tag. Windmill v1.786.1 avoids content shift on home page load and in the script editor logs pane. ms-swift v4.4.3 fixes a batch_sampler set_epoch bug. Nav2 1.5.1 is tagged but no detailed changelog given.
GLM 5.2 from Z.AI reduced input price by 20% to $0.40 per 1M tokens. DeepSeek V4 Pro from DeepSeek raised input price by 43% to $0.63 per 1M tokens. Qwen3.5 397B A17B from Alibaba raised input price by 28% to $0.50 per 1M tokens.
Two projects were detected today. TradingAgents-astock (2,801 stars) is a multi-agent framework for A-share investing, adapting the TradingAgents architecture to Chinese stock market data sources. linkedin-skills (534 stars) provides 11 Claude Code and Codex skills for generating human-sounding LinkedIn posts and comments.
ComBodied Agents by Ding, Wang, Feng et al. proposes a paradigm where software agents handle communication and embodied agents handle physical action, working together for tasks like medication adherence. The paper highlights that a single agent type cannot explain human context such as forgetting, confusion, side effects, or refusal.
The day's activity shows steady project maintenance, selective price adjustments by model providers, and the emergence of specialised agent frameworks for niche markets. The high star count reflects continued community interest in new tools and skills.
Source: olud.ai tracking of 10,000+ open-source AI projects, 300+ models and live provider pricing. Figures are measured, not estimated. All releases · Live pricing · Latest in AI
