agents-libagents-lib
AgentsMCP ServersWorkflowsTutorialsComparisonsTemplatesNewsAbout
AiriAI Companionslast30days-skillImpeccableag-kitHermes WebUIClaude VideoGitHub TrendingNews RoundupAgent Research

AI News Roundup: July 27, 2026 — Airi Brings AI Companions to Self-Hosted, Agent Skills Reach 54K, and Design Intelligence Goes Mainstream

Airi: Self-Hosted AI Companions Reach 43,700 Stars

moeru-ai/airi has surged to 43,700+ GitHub stars, establishing itself as the leading self-hosted AI companion platform. Inspired by the Neuro-sama phenomenon, Airi lets users run their own AI waifu/virtual character with real-time voice chat and game integration, all fully user-owned.

What makes it stand out:

  • Cross-platform: PWA in browsers, Electron desktop (Windows/macOS/Linux), and Capacitor mobile
  • Game integration: Plays Minecraft, Factorio, Kerbal Space Program, and Helldivers 2 alongside users
  • Real-time voice chat with Discord integration for seamless community interaction
  • Memory system powered by DuckDB WASM, giving companions persistent identity
  • VRM and Live2D model support with full animation pipelines
  • Multi-provider TTS: ElevenLabs, Azure, and local Kokoro TTS options
  • Multi-LLM backend: OpenAI, Anthropic Claude, DeepSeek, and OpenRouter via the xsai library

Why it matters: Airi represents a new category — fully user-owned AI companions that combine voice, gaming, memory, and avatars. Unlike cloud-bound chatbots, Airi runs locally and integrates deeply with gaming ecosystems, signaling a shift toward agentic presence in entertainment.

last30days-skill: The Agent-Led Search Engine Hits 54K Stars

mvanhorn/last30days-skill has topped GitHub Trending at 54,000+ stars, redefining how AI agents research the internet. The skill searches 18+ platforms in parallel — Reddit, X/Twitter, YouTube, TikTok, Hacker News, Polymarket, GitHub, LinkedIn, Bluesky, Perplexity and more — then synthesizes a grounded, people-scored summary.

Key innovations:

  • People-first scoring: Ranks results by real engagement (upvotes, likes, views, real money) instead of SEO or editorial curation
  • Full content depth: Pulls YouTube transcripts, Reddit top comments, TikTok captions — not just titles
  • Parallel cross-source analysis: Merges the same story across platforms into unified clusters
  • Multiple modes: People, company, topic, and comparison research
  • Structured outputs: Markdown, JSON, and HTML briefs available

Why it matters: This is the most ambitious agent-led research tool to date. At 54K stars and ranked #1 trending, it signals that AI agents are moving beyond code into genuine intelligence work — research, synthesis, and competitive analysis.

Impeccable: AI Coding Agents Get a Design Language at 51K Stars

pbakaus/impeccable has reached 51,100+ stars, offering a design guidance system built specifically for AI coding agents. With 23 design commands, live browser iteration, and 60 deterministic anti-pattern detectors, it gives agents an explicit design vocabulary.

Key features:

  • 23 design commands: /impeccable polish, /impeccable audit, /impeccable critique, /impeccable animate, /impeccable bolder, /impeccable quieter and more
  • 60 deterministic detectors that run without LLMs to catch AI-generated frontend anti-patterns
  • Design hooks for Claude Code, GitHub Copilot, Codex, Cursor, Grok Build, and more
  • Supported tools: Cursor, Claude Code, GitHub Copilot, Gemini CLI, Codex, Grok Build, OpenCode, Pi, Kiro, Trae, Rovo Dev, Qoder, Mistral Vibe

Why it matters: AI coding agents have long been criticized for generating visually mediocre UI. Impeccable is a direct response — it codifies design principles into commands agents can actually execute, closing the gap between code quality and visual polish.

ag-kit: The Antigravity-First Agent Engineering Kit

vudovn/ag-kit is a newer entrant at 7,800+ stars, positioning itself as a "抗引力优先" (antigravity-first) agent engineering framework. It combines rules, skills, specialist agents, workflows, persistent memory, MCP guidance, and a native safety hook.

Key features:

  • 20 domain specialist agent roles with built-in definitions
  • 47 progressive skills with validation helpers
  • 13 repeatable workflow slash-commands
  • 6 workspace-wide rules for routing and safety
  • Native PreToolUse gate that blocks destructive commands like rm -rf / while allowing safe cleanup
  • Merge-aware updates with rollback support
  • Plugin bundling with SHA-256 content inventory

Why it matters: ag-kit takes a security-first approach to agent frameworks, with the PreToolUse gate being a standout feature. As agents get more powerful tool access, built-in safety becomes table stakes.

Hermes WebUI: The Hermes Agent Gets a Full Web Experience

nesquena/hermes-webui has reached 16,600+ stars, providing a full-featured web interface for the Hermes Agent. With a three-panel layout (chat, sessions, files), it delivers full CLI parity from any browser or phone.

Key features:

  • Streaming responses, session management, voice input, and profile switching
  • Mobile-responsive design with customizable themes
  • Optional password auth and passkeys for security
  • Built with vanilla Python and JS — no build step, no framework, no bundler
  • Docker and Nix deployment options available

Claude Video: Giving Agents Vision for 400+ Video Platforms

bradautomates/claude-video has reached 10,600+ stars, enabling Claude to actually watch and understand video content. The /watch command downloads videos, extracts frames, transcribes audio, and feeds everything to the agent.

Why it matters: Video understanding was a major blind spot for AI agents. claude-video closes it — supporting YouTube, TikTok, Instagram, Vimeo, Loom, X, and 400+ platforms with four detail modes and frame deduplication.

Claude Cookbooks: Anthropic's Official Examples Hit 50K

anthropics/claude-cookbooks has crossed the 50,000 star milestone — a strong signal that Anthropic's agent ecosystem has achieved mainstream developer adoption.

The Bigger Picture

July 27, 2026 brings four major themes:

  1. AI companions are going self-hosted. Airi's 43K stars show that users want ownership over their AI personalities — not just chat, but voice, gaming, and persistent identity.

  2. Agent-led research is hitting critical mass. last30days-skill at 54K and topping GitHub Trending signals that AI research agents are graduating from prototype to daily driver.

  3. Design intelligence becomes a framework capability. Impeccable at 51K proves there's massive demand for giving AI coding agents explicit design vocabulary — not just code generation, but visual quality.

  4. Security-first agent frameworks emerge. ag-kit's PreToolUse safety gate represents a new direction: frameworks that assume agents will make mistakes and build guardrails in.

Sources