AI Open Source Radar — August 10, 2026

Selection basis: Trendshift’s daily “stars today” (data confirmed fresh) is the primary metric, filtering for repos with the steepest gains in the last 24 hours. Large, mature, or plateaued repos — and yesterday’s picks, which all dropped off the daily board today — are excluded, so a short list of genuine breakouts is the norm.

msitarzewski/agency-agents — ~+648 ⭐ today (70 mentions, #1 daily)

Cumulative: surging · github.com/msitarzewski/agency-agents

A catalog-style open-source project bundling 230+ specialized AI agent “personas” — from frontend wizards to Reddit community experts to reality-checkers. Each agent is designed as a domain expert with its own voice, work process, and concrete deliverables (code, checklists, and so on) rather than a generic prompt template, and the roster is organized into divisions by domain. Agents install directly into real coding tools such as Claude Code, Cursor, Codex, Gemini, and Osaurus, and a native app (agency-agents-app) for macOS, Linux, and Windows lets you browse, install, and track them. Born from a Reddit thread and refined over months, it stands out for standardizing reusable “expert role” assets — and its social mentions spiked sharply today.

Specialized expertise per agent — not generic templates
Personality-driven, with a distinct voice and communication style
Deliverable-focused, with real code, processes, and measurable outcomes
Installs into Claude Code, Cursor, Codex, Gemini, Osaurus, and more

Practical use: standardize a team’s repetitive work (review, QA, documentation) as role-specific agents and deploy them consistently across multiple coding tools.

Tags: #AgentPersonas #ClaudeCode #Cursor #CodexCLI #DevWorkflow #AgentCatalog

ZhuLinsen/daily_stock_analysis — ~+4,391 ⭐ today (58 mentions, #2 daily) [renewed surge]

Cumulative: ~50k+ · github.com/ZhuLinsen/daily_stock_analysis

An LLM-powered, multi-market stock-analysis automation system covering A-shares, Hong Kong, and US markets. Using free data sources such as AkShare, Baostock, and YFinance, it collects quotes, real-time news, technical indicators, sentiment signals, and fundamentals, then has an LLM synthesize them into a daily decision dashboard. Results are pushed automatically to WeChat Work, Feishu, Telegram, Discord, Slack, or email, and it can run unattended and cost-free on a schedule via GitHub Actions. It supports multiple LLM providers and both local and Docker deployment. Already a large project with roughly 50k stars, it re-entered the top of the board today with a steep daily gain (#2) — included as an exception precisely because its momentum is re-accelerating rather than slowing.

Markets: A-shares, Hong Kong, US
Free data sources: AkShare, Baostock, YFinance
Push channels: WeChat Work, Feishu, Telegram, Discord, Slack, email
Scheduled, cost-free unattended runs via GitHub Actions; local or Docker deploy

Practical use: a reference architecture for research and asset-management teams to automate ticker monitoring and a daily briefing on a free schedule.

Tags: #StockAnalysis #DecisionDashboard #GitHubActions #MultiMarket #AlertPush #LocalLLM

stablyai/orca — ~+440 ⭐ today (26 mentions, #15 daily, new)

Cumulative: surging · github.com/stablyai/orca

An “Agent Development Environment (ADE)” for running a fleet of coding agents in parallel. It runs 30+ coding agents (Claude Code, Codex, OpenCode, and others) side by side, each in an isolated git worktree, with terminal splits, an embedded Chromium browser, SSH remotes, and GitHub/Linear integration. Because you bring your own CLI subscriptions and model credits (BYO), the platform itself is MIT-licensed and free, and runs on desktop, mobile, and VPS. Differentiators include a Design Mode that lets you click a UI element in the embedded Chromium and send the HTML/CSS and a screenshot into an agent’s prompt, plus scriptable worktree create, snapshot, click, and fill commands. Backed by Y Combinator, it launched recently and is climbing fast.

Runs 30+ coding agents in parallel in isolated git worktrees
Terminal splits, embedded Chromium, SSH remotes, GitHub/Linear integration
BYO subscriptions and credits; platform is MIT-licensed and free
Design Mode plus scriptable worktree/click/fill automation

Practical use: build a “parallel coding” workflow that delegates one task to several agents at once, then compares and merges the results.

Tags: #AgentOrchestration #GitWorktree #ParallelAgents #EmbeddedChromium #BYOK #CodingADE

ShawnPana/phone-harness — ~+274 ⭐ today (23 mentions, #10 daily, new)

Cumulative: surging · github.com/ShawnPana/phone-harness

A tool that lets an AI agent operate a real iPhone. It uses iPhone Mirroring (macOS Sequoia and later) to render the phone as a Mac window, then forwards real mouse and keyboard input as touches. It captures the mirroring window and uses OCR via Apple’s Vision framework to identify every visible string with tap-ready coordinates, enabling taps, long-presses, drags/flicks, scrolls, Unicode typing, and app shortcuts. With no separate agent or automation SDK — just built-in mirroring and system permissions (Accessibility and Screen Recording) — it can drive the phone UI, which makes it useful for mobile app testing and repetitive-task automation. Its novel approach drove the recent surge.

Renders the iPhone via iPhone Mirroring; forwards mouse/keyboard as touches
Apple Vision OCR maps every on-screen string to tap-ready coordinates
Supports taps, long-presses, drags/flicks, scrolls, Unicode typing, app shortcuts
Needs only iPhone Mirroring plus Accessibility and Screen Recording permissions

Practical use: let QA and ops teams automate repetitive iPhone-app interactions and regression tests with an agent, without wiring up a physical device rig.

Tags: #iPhoneMirroring #MobileAgent #VisionOCR #UIAutomation #macOS #TouchControl

vitali87/code-graph-rag — ~+328 ⭐ today (18 mentions, #5 daily)

Cumulative: ~3k+ · github.com/vitali87/code-graph-rag

A “code knowledge-graph RAG” tool built for monorepos. It parses a multi-language codebase with Tree-sitter to build an AST, then loads that into a Memgraph graph database as a structured knowledge graph. Natural-language queries flow as: the AI generates a Cypher query, runs it against the graph, and returns a response — letting you query, understand, edit, and optimize code structure in plain English. It supports Python, TypeScript, JavaScript, Rust, Go, Java, C/C++/C#, PHP, Lua, and Dart under one unified schema, and adds intent-based search via UniXcoder embeddings so you can find code by describing what it does (e.g., “error-handling functions”). A key differentiator is that it runs as an MCP server, so you can query and edit the codebase directly from Claude Code.

Tree-sitter parsing → AST → Memgraph knowledge graph
Natural language → Cypher → graph results → response
Unified schema across Python, TypeScript, JS, Rust, Go, Java, C/C++/C#, PHP, Lua, Dart
Intent-based search via UniXcoder embeddings; runs as an MCP server

Practical use: in large, multi-language monorepos, precisely condense the context handed to an AI via the graph to improve accuracy and cut token cost.

Tags: #KnowledgeGraph #TreeSitter #Memgraph #CypherQuery #MonorepoRAG #MCP