Intelligence Digest

OpenHuman, an open-source personal AI assistant built in Rust, gained 84 stars and 14 forks in the past 24 hours on GitHub, with 26 pushes and 5 pull requests indicating active development. The project addresses growing demand for private, locally-run AI assistants that don't rely on cloud services, offering users full control over their data while delivering powerful agentic capabilities. OpenHuman is designed as a desktop-first, UI-focused agent with simple onboarding, persistent memory, and integrations with productivity tools, distinguishing itself from config-heavy or terminal-based alternatives.

The GitHub repository colbymchenry/codegraph gained 273 stars and 8 forks in the past 24 hours, introducing a pre-indexed code knowledge graph for Claude Code that reduces token usage and tool calls by operating 100% locally. This project addresses a real pain point in AI-assisted development by significantly reducing token usage and tool calls, which directly lowers costs and improves efficiency for developers using Claude Code. The tool is written in TypeScript and operates 100% locally, meaning no external API calls are needed for code graph operations, which enhances privacy and reduces latency.

The GitHub repository rohitg00/ai-engineering-from-scratch gained 86 stars and 15 forks in the past 24 hours, signaling rapid community adoption of its hands-on AI engineering curriculum. This trending repository fills a critical gap in AI education by teaching engineers to understand and build AI systems from mathematical foundations rather than just calling APIs. The curriculum spans 416 lessons across 20 phases, covering everything from linear algebra to autonomous multi-agent swarms, with all code written in Python.

The rohitg00/agentmemory repository on GitHub has gained 61 stars in the past 24 hours, offering a TypeScript-based persistent memory solution for AI coding agents grounded in real-world benchmarks. As AI agent-based systems become more prevalent, persistent memory is a critical challenge, and this project addresses a timely need for coding assistants to retain context across sessions. The repository is written in TypeScript, has gained 5 forks and 4 pushes with 1 pull request, and claims to be the

The withcoral/coral repository gained 40 stars in 24 hours, introducing a local-first SQL runtime that lets agents and developers query APIs, files, and live data sources through a single SQL interface built in Rust. Coral addresses a key pain point in data integration by providing a composable, single SQL interface for agents, eliminating the need for bespoke tool glue when working with heterogeneous data sources, making it highly relevant for data engineers and agent-based system developers. Coral is implemented in Rust, supports querying from the CLI, allows schema and table inspection, and can expose its runtime over MCP (Model Context Protocol) for agent integration, with 90 pushes and no pull requests in the analyzed period.

Anthropic has launched an official, company-managed directory for high-quality Claude Code Plugins on GitHub, gaining 35 stars and 5 forks within 24 hours. This signals Anthropic's structured approach to extending Claude's capabilities through plugins, indicating a maturing ecosystem for Claude Code. The repository is written in Python and includes 2 pushes and pull requests, suggesting active development.

rtk-ai/rtk is a Rust-based CLI proxy that reduces LLM token consumption by 60-90% on common development commands, gaining 35 stars in 24 hours. It operates as a transparent proxy between your LLM agent and command-line tools, compressing command outputs before they reach the AI context window. This tool addresses the hidden cost of token bloat in AI-assisted development, enabling longer coding sessions and lower API costs without any configuration changes. For developers using AI coding agents, this directly translates to significant cost savings and improved efficiency. RTK is a single Rust binary with zero dependencies, adding less than 10ms overhead while supporting over 100+ commands. It compresses outputs like git diff from 10,000 tokens to 2,500 and pytest output from 8,000 to 800, demonstrating substantial token reduction across common development workflows.

The GitHub repository Imbad0202/academic-research-skills, a Python-based toolkit providing academic research skills for Claude Code, gained 155 stars in the past 24 hours, covering a full workflow from research to writing, reviewing, revising, and finalizing. This repository addresses a growing demand for AI-assisted academic research automation, offering researchers and students a structured workflow tool that integrates directly with Claude Code to streamline the end-to-end research process. The repository is written in Python and has seen minimal community engagement so far, with 0 forks and only 2 pull requests despite the rapid star growth, suggesting it is still in early adoption stages.

The GitHub repository alchaincyf/nuwa-skill gained 38 stars in 24 hours for its Python-based tool that aims to distill cognitive models, decision-making heuristics, and 'expression DNA' from individuals, with the tagline 'The next employee you want to distill doesn't have to be a colleague.' This project touches on a novel intersection of knowledge management and AI-assisted learning by attempting to capture and replicate expert thinking processes, which could transform how organizations preserve and transfer tacit knowledge. The repository is written in Python and focuses on extracting mental models, decision heuristics, and expression patterns from individuals, though the description remains vague with limited technical depth or clearly defined use cases.

The Yuan1z0825/nature-skills repository is a trending GitHub project that gained 34 stars in the past 24 hours, offering Python-based skills for academic writing and scientific plotting that conform to Nature journal standards. This tool addresses a real pain point for academic researchers who struggle to meet the high publication standards of top-tier journals like Nature, potentially saving significant time and effort in manuscript preparation. The toolkit includes two main skills: 'nature-figure' for creating publication-quality figures and 'nature-polishing' for refining academic writing, both grounded in primary Nature papers.

Thysrael/Horizon, an open-source AI-powered news radar built in Python, gained 32 stars and 4 forks on GitHub in the past 24 hours. The tool aggregates content from Hacker News, Reddit, RSS, Telegram, and GitHub, then generates personalized daily briefings in both English and Chinese. Horizon addresses a growing need for personalized, multilingual news curation in an era of information overload, offering users control over trusted sources, filtering thresholds, and delivery channels. Its open-source nature and support for both English and Chinese make it accessible to a broad international audience. Horizon supports customizable sources, AI scoring and deduplication, background context enrichment, comment summaries, and a community source hub, with briefings deliverable via email, Feishu, Slack, or GitHub Pages, and deployable via Docker and GitHub Actions.