Today's trending repositories tell a coherent story: the agentic era is moving toward specialization. Rather than general-purpose chat interfaces, developers are building agents tuned for specific contexts — financial workflows, research tasks, multi-model routing — and investing in the shared infrastructure that makes those agents composable and resilient across different providers.
DeepSeek-TUI: A Terminal Coding Agent Built in Rust
Hmbown/DeepSeek-TUI is a Rust-based coding agent designed to run entirely inside a terminal, powered by DeepSeek models, and it surged to 3,731 new stars today. The project positions itself as a lean alternative to GUI-heavy coding assistants — keeping the development loop fast and keyboard-driven, with no browser or Electron runtime required. For developers who prefer staying in the terminal, a capable Rust-compiled agent that integrates tightly with the shell is a meaningful option beyond the current crop of IDE-embedded tools.
github.com/Hmbown/DeepSeek-TUI
Anthropic Financial Services: Claude Tooling for Finance
anthropics/financial-services earned 3,660 stars today as a collection of tools and applications applying Claude to financial services use cases. The repo bundles ready-made patterns for tasks like document analysis, data extraction from financial reports, and client-facing summarization. Given the regulatory sensitivity of finance, having an official Anthropic-maintained reference implementation signals that industry-specific agentic tooling is becoming a first-class concern, not just a community experiment.
github.com/anthropics/financial-services
Agent Skills: A Production Engineering Skill Library for AI Agents
addyosmani/agent-skills collected 1,893 stars today with a focused promise: a curated set of production-grade engineering skills for AI coding agents. The Shell-based library packages reusable agent behaviors — running tests, linting, navigating monorepos, parsing CI output — so teams don't need to re-prompt their way to the same patterns on every project. As AI coding agents get embedded in real engineering workflows, shared skill libraries like this could become as ubiquitous as shared component libraries are in frontend development today.
github.com/addyosmani/agent-skills
9router: Multi-LLM Routing with Auto-Fallback
decolua/9router grabbed 1,052 stars today as a routing layer that connects Claude, GitHub Copilot, and other AI tools to multiple LLM backends with automatic fallback. When a primary model is unavailable or hits rate limits, 9router quietly switches to an alternate provider without any changes to client code. As teams rely on AI tools throughout the development cycle, this kind of resilience layer starts to look less like a nice-to-have and more like basic infrastructure.
Local Deep Research: Offline Research at 95% SimpleQA Accuracy
LearningCircuit/local-deep-research reached 559 new stars today, offering a privacy-preserving research tool that runs entirely on local hardware against multiple search engines and local LLMs. The project claims around 95% accuracy on the SimpleQA benchmark — competitive with hosted alternatives — while keeping all queries and results off third-party servers. For developers or researchers working with sensitive data or simply wanting to avoid routing every search through an external API, this fills a real gap in the local-AI stack.