AI News Roundup: July 25, 2026 — Block Buzz Launches, Alibaba Open Code Review Goes Open Source, and Ego Lite Redefines Agent Browser

Block Launches Buzz: A Self-Hostable Workspace Where Humans and AI Agents Share the Same Room

Block, Inc. has released Buzz, a self-hostable communication platform built on Nostr protocol technology that enables humans and AI agents to collaborate in shared workspaces. With over 12,000 GitHub stars and 2,400+ stars on launch day, Buzz has immediately captured attention as a bold new paradigm for agent-human collaboration.

Why this matters: Buzz is not a chat app with bot integration. It's a unified event log where every message, reaction, workflow step, and git event exists as a cryptographically signed Nostr event. Agents participate as equal members — not automated bots — with their own keys, memberships, and audit trails. The "branch-as-room" concept means feature branches become channels where patches, CI results, code reviews, and merge decisions coexist in the same space as the conversation that spawned them.

The bigger story: Block's entry into the agent collaboration space signals that the financial services industry sees the agent paradigm as infrastructure, not a novelty. With ACP harness support for Goose, Codex, and Claude Code, Buzz is designed to interoperate with the existing agent ecosystem rather than replace it.

Alibaba Open Source Its Internal AI Code Review Tool — Battle-Tested at Scale

Alibaba Group has open-sourced its internal AI-powered code review CLI tool, which has served tens of thousands of developers and identified millions of code defects over two years of internal use. The tool, named Open Code Review (OCR), is built on a hybrid architecture combining deterministic rule pipelines with LLM Agent capabilities.

Why this matters: This is not a toy project. OCR has been production-tested at one of the world's largest e-commerce platforms. Its fine-tuned ruleset covers NPE, thread-safety, XSS, and SQL injection — the kinds of vulnerabilities that cause real production incidents. The hybrid architecture (deterministic rules + LLM) is a pragmatic middle ground between purely rule-based linters and purely LLM-based reviewers.

Key technical detail: OCR supports both OpenAI and Anthropic-compatible models, making it model-agnostic in a way that most code review tools are not. It also offers a "delegation mode" where AI coding agents can perform reviews using their own LLM — a novel approach to agent self-review.

Ego Lite: The Fastest Browser for AI Agents to Run Web Automation

Citrolabs has released Ego Lite, a browser designed specifically for AI agents to perform web automation while sharing your logged-in browser state — without disrupting your own browsing. With nearly 4,000 GitHub stars and 980+ stars per day, Ego Lite has quickly become the most compelling browser tool for AI agents in 2026.

Why this matters: Ego Lite's key innovation is its JavaScript-function-based approach instead of CLI commands, enabling 2.5× faster execution with fewer tokens. The dedicated "Spaces" feature gives each agent an isolated workspace while inheriting your Chrome logins, cookies, and extensions on first launch. Deeply nested iframes that break other browser tools are handled via kernel-level customization.

Performance: Benchmarks show Ego Lite finishing tasks up to 2.5× faster than Vercel's agent-browser, with the gap widening as task complexity increases. The project is free, open-source (MIT), and compatible with any agent (Claude Code, Codex, Cursor, etc.) via the ego-browser skill.

ECC: The Agent Harness Operating System Hits 230K+ Stars

affaan-m/ECC has surged past 230,000 GitHub stars, establishing itself as the dominant agent harness optimization system. ECC provides cross-platform support for Claude Code, Codex, Cursor, OpenCode, Gemini, Zed, and GitHub Copilot, along with 67 specialized subagents, 279 workflow skills, and an AgentShield security scanner with 1,282 tests.

Why this matters: ECC represents the maturation of the "agent harness" category — the layer between the raw LLM and the developer workflow. With 230+ contributors and continuous learning v2, ECC is becoming the de facto operating system for agentic development across the major coding platforms.

GitHub Trending: Buzz, Ego Lite, and ECC Dominate

This week's GitHub trending is dominated by agent infrastructure and collaboration tools:

Repository	Stars	Daily Gains	Description
Buzz	12,278	2,491	Self-hostable human-AI workspace on Nostr
Open Code Review	13,166	431	Alibaba's hybrid AI code review CLI
Ego Lite	3,787	986	Fastest browser for AI agent web automation
ECC	233,413	377	Agent harness optimization system
mattpocock/skills	188,616	1,740	Real engineer skills for Claude Code
awesome-claude-skills	70,723	577	Curated Claude Skills collection
Harper	13,507	503	Offline grammar checker
Chat2DB	26,789	360	AI-driven database SQL client

Notable: The skills ecosystem (mattpocock/skills, awesome-claude-skills) continues its explosive growth, while agent collaboration tools (Buzz) and agent browser tools (Ego Lite) emerge as new categories.

Other Notable Trends

aisuite: Andrew Ng's Unified AI Provider Interface

andrewyng/aisuite — a simple, unified interface to multiple Generative AI providers — has reached 15,200+ stars. As model proliferation continues (Kimi K3, Gemini, Claude 4.7, GPT-5.6, DeepSeek R1/R2), tools that abstract provider-specific APIs become increasingly valuable.

TurboVec: Rust-Native Vector Index with Python Bindings

RyanCodrai/turbovec — a vector index built on TurboQuant, written in Rust with Python bindings — has reached 14,300+ stars. Rust-native vector stores are gaining traction as developers seek alternatives to Python-heavy RAG stacks.

Agent-Specific Infrastructure Proliferation

The past week saw the launch of several agent-specific infrastructure tools:

Buzz — Agent collaboration workspace (Block)
Ego Lite — Agent browser with logged-in state sharing
Open Code Review — AI-powered code review CLI
ECC — Agent harness optimization system

The Bigger Picture

Three themes dominate July 25, 2026:

Agent collaboration becomes a category. Buzz is the first serious attempt to create a workspace where humans and agents have equal footing — not just integration, but participation. The Nostr-based architecture is elegant: every interaction is a signed event in a unified log, giving both humans and agents the same affordances and audit trails.
Agent tooling shifts from "what can agents do?" to "how well do agents do it?". Open Code Review's hybrid architecture (deterministic + LLM), Ego Lite's 2.5× performance over agent-browser, and ECC's 279 workflow skills all reflect a maturing ecosystem focused on quality, speed, and reliability — not just capability.
The skills ecosystem is becoming infrastructure. With mattpocock/skills at 188K stars and awesome-claude-skills at 70K, skills are no longer a niche — they're the de facto standard for extending agent capabilities. This has implications for every agent framework: the ones that integrate with the skills ecosystem will win.