Agent Skills, Vector Search, and Offline AI Heat Up Together

Agent momentum has not faded, but today’s open source attention is shifting more clearly toward skills, executable capability, and output quality; at the same time, vector indexing, knowledge bases, local inference, offline AI, and in-database durable execution form a new set of infrastructure signals.

Hot Projects

1. RyanCodrai/turbovec: a vector index built on TurboQuant, written in Rust with Python bindings, gaining 1,554 new stars today and ranking as the highest-growth project
2. NousResearch/hermes-agent: an Agent system that grows with the user, gaining 1,112 new stars today and showing Agent core systems still have strong momentum
3. mvanhorn/last30days-skill: an Agent Skill for researching any topic across Reddit, X, YouTube, HN, Polymarket, and the web, gaining 1,111 new stars today and heating up sharply from yesterday
4. Leonxlnx/taste-skill: a skill intended to give AI “better taste” and reduce boring, generic, templated output, gaining 1,103 new stars today
5. lfnovo/open-notebook: an open source implementation of Notebook LM, gaining 554 new stars today and continuing to represent knowledge organization and notebook workflows
6. TapXWorld/ChinaTextbook: a collection of PDF textbooks for primary school, secondary school, high school, and university, gaining 350 new stars today and reflecting demand around digitized education resources
7. aaif-goose/goose: an open source AI Agent that can install, execute, edit, and test, gaining 322 new stars today and representing Agents moving from suggestions toward real task execution
8. microsoft/pg_durable: PostgreSQL in-database durable execution, gaining 316 new stars today and standing out as a backend infrastructure project worth watching
9. Crosstalk-Solutions/project-nomad: a self-contained offline survival computer project that integrates key tools, knowledge, and AI, gaining 309 new stars today
10. openai/plugins: the OpenAI Plugins project, gaining 262 new stars today and showing continued residual attention around plugins and tool-calling ecosystems
11. refactoringhq/tolaria: a desktop application for managing Markdown knowledge bases, gaining 245 new stars today
12. yikart/AiToEarn: a project centered on “making money with AI,” gaining 183 new stars today and representing continued attention around AI commercialization narratives
13. ggml-org/llama.cpp: a C/C++ local LLM inference project, gaining 158 new stars today and continuing to represent mature local model runtime infrastructure
14. opencv/opencv: the classic open source computer vision library, gaining 65 new stars today and looking more like long-term stable infrastructure than a breakout daily hotspot
15. HunxByts/GhostTrack: a location or phone-number tracking tool, gaining 28 new stars today and belonging to the category of projects that should be marked for privacy and compliance risk

Trend

1) Agent shifts from product integration toward skills, execution, and quality control

• hermes-agent gained 1,112 new stars today, last30days-skill gained 1,111, taste-skill gained 1,103, and goose gained 322, showing the Agent mainline remains strong.
• Compared with yesterday’s CopilotKit, MemPalace, Agent-Reach, and superpowers, today’s Agent hotspots lean more toward skill packaging, output quality, and real execution capability.
• This shows Agent deployment is splitting into two practical questions beyond product integration: how to give Agents reusable skills, and how to make their output more judgment-driven, less templated, and capable of completing real tasks such as installation, editing, and testing.

2) AI output quality becomes a new branch of Agent Skills

• taste-skill gained 1,103 new stars today, making it one of the most notable new projects.
• Its focus is not on giving AI more tools or expanding context, but on reducing dull, generic, boring, and templated output.
• This suggests a new branch may be forming within Agent Skills: moving from “make AI do more” toward “make AI do better.” For content generation, product copy, code review, research summaries, and automation workflows, quality control and taste may become as important as tool calling.

3) Vector indexing and AI data retrieval heat up again

• RyanCodrai/turbovec gained 1,554 new stars today, making it the highest-growth project on the list.
• open-notebook gained 554 new stars, while tolaria gained 245, showing continued demand around knowledge organization, Markdown knowledge bases, and notebook workflows.
• Viewed together, turbovec, open-notebook, and tolaria show AI-ready data is no longer just about OCR or document conversion. It is expanding toward “vector indexing + knowledge management + reusable workflows.” For RAG, research assistants, enterprise knowledge bases, and personal knowledge systems, retrieval performance and higher-level organization will both matter.

4) Local inference and offline AI reappear

• llama.cpp gained 158 new stars today, while project-nomad gained 309, representing mature local LLM inference and self-contained offline AI systems.
• This line does not conflict with cloud Agents. It emphasizes low network dependence, personal control, offline availability, and information access in special environments.
• As Agents and knowledge bases depend more on external services, local inference, offline knowledge packs, self-contained toolkits, and low-resource runtime capability will remain a long-term demand rather than merely a nostalgic local-first route.

5) Durable execution starts moving into the database layer

• microsoft/pg_durable gained 316 new stars today and brings durable execution into PostgreSQL itself, making it one of today’s most important infrastructure signals.
• The past few days focused more on Agent core systems, context, knowledge input, frontend integration, and tool calling. pg_durable pulls attention back toward reliable execution, transaction boundaries, asynchronous tasks, and backend workflows.
• If Agent long-running tasks, automation processes, and business systems continue to converge, reliable scheduling, state persistence, failure recovery, and observable execution will become more important. In-database durable execution may become a new example alongside workflow engines and backend task systems.

6) Education resources and AI commercialization narratives still draw attention, but need careful interpretation

• ChinaTextbook gained 350 new stars today, showing clear demand around education resources, PDF content libraries, and study material organization.
• AiToEarn gained 183 new stars today, representing the continued appeal of “making money with AI” narratives.
• Both categories reveal demand, but neither should be treated as a directly repeatable product opportunity. Education-resource projects require attention to copyright boundaries, while AI income projects need to be separated into real tools, content aggregation, and short-term traffic narratives.

7) Traditional infrastructure and risk-heavy tools remain peripheral signals

• opencv/opencv gained 65 new stars today, reflecting stable attention around mature computer vision infrastructure rather than a daily breakout.
• GhostTrack gained 28 new stars today and involves location or phone-number tracking, with obvious privacy, compliance, and abuse risks.
• These projects do not form today’s mainline, but they are useful reminders that Trending includes both long-term infrastructure and risk-heavy tools. It is reasonable to record their appearance, but high-risk tools should not be treated as positive product references.

Today’s Judgment

The most important judgment today is that the Agent mainline continues, but its internal focus has shifted from productized components toward skills, execution capability, and output quality; meanwhile, the AI data layer and reliable backend execution are becoming more visible again.

last30days-skill, taste-skill, hermes-agent, and goose show Agents remain a core hotspot, but today’s keywords are no longer only Memory, Generative UI, and internet access. They are research skills, taste, executable tasks, and Agent core systems. turbovec, open-notebook, and tolaria show knowledge retrieval and knowledge management continue to strengthen. llama.cpp and project-nomad show local and offline AI remain long-term needs. pg_durable suggests durable execution may become a new infrastructure direction for Agent long-running tasks and backend automation.

Tomorrow, the three things to watch are whether turbovec stays hot, which would confirm vector indexing as a new infrastructure hotspot; whether taste-skill, last30days-skill, and goose continue, validating momentum around Agent Skills and executable Agents; and whether pg_durable remains on the list, showing whether durable execution can expand from a single project into a clearer backend trend.