PullRepo

Daily radar for the fastest-growing AI tools & repos

Today's Code Assistant: Fastest-Growing Projects — July 22, 2026

Today's the Code Assistant space, we see a variety of projects gaining traction across different platforms and use cases, ranging from innovative OCR solutions to desktop widgets for quota tracking. Among them, Jia-Ethan/codex-keysmith stands out with its high star count and steady growth, demonstrating strong community interest in version-independent Codex instruction deployment.

arcships/light-ocr is a fast, offline Optical Character Recognition (OCR) solution designed for Node.js and C++. It leverages PP-OCRv6 with Core ML/WebGPU hardware acceleration to recognize text within images, providing confidence scores and coordinates. This project has seen significant growth over the past month, likely due to its efficient performance and broad compatibility across multiple platforms.

Jia-Ethan/codex-keysmith offers version-independent deployment of Codex instructions with features such as dry-run capabilities, backups, hook isolation, and recovery mechanisms. With a substantial 1,262 stars on GitHub and a solid growth score, this tool is gaining popularity for its comprehensive set of utilities that enhance the reliability and flexibility of Codex-based systems.

CodeDrobe/skills provides an open-source theming skill for OpenAI Codex, including AI-generated themes and cross-platform runtime support. Although it has fewer stars compared to some other projects in this list (221), its steady growth suggests a niche but growing community interested in customizing their Codex desktop experience.

PKU-YuanGroup/OpenAI4S is described as a replication of Claude Science using 9.9 元豆包API, though the exact functionalities are not clear from the provided description. Despite having fewer stars (125), it has seen considerable development activity with over 100 commits in the last month, indicating ongoing project momentum and potential interest within specific user communities.

worldwonderer/novel-to-game is a pipeline designed to turn any novel into a playable game through a series of seven skills adapted for Claude Code, Codex, and Kimi Code. With 92 stars, this tool reflects an interesting intersection between literature and gaming, offering creators a way to bring their written stories to life in interactive formats.

shanggqm/codexU is a macOS desktop widget that tracks OpenAI Codex usage, quota limits, token consumption, and provides daily task boards. Its 287 stars indicate strong interest from the developer community for tools that help manage API quotas more effectively, especially as Codex becomes increasingly integrated into workflows.

HEXUXIU/M365-Copilot2API enables access to Microsoft 365 Copilot via a command-line interface and an OpenAI-compatible API. Despite having fewer stars (71), the project's steady growth score suggests it is filling a niche need for developers looking to interact with M365 Copilot through more familiar APIs.

KilimcininKorOglu/M365Bridge offers a Go implementation that converts Microsoft 365 Copilot’s WebSocket interface into an OpenAI/Anthropic compatible HTTP API, supporting tool calls. With only 33 stars but robust development activity (100 commits in the last month), this project is likely addressing specific integration challenges faced by developers working with M365 Copilot.

videlalvaro/inference-school provides a hands-on Swift and Metal course for building large language model (LLM) inference on Apple silicon, complete with 48 guided lessons, runnable exercises, and a companion book. Attracting 176 stars, this educational resource is growing steadily as more developers seek to understand the technical underpinnings of LLMs.

QuintinShaw/openasr focuses on local-first speech-to-text solutions, offering seven model families, signed model catalogs, and an OpenAI-compatible API. With a notable 116 stars, this project appeals to users who prioritize privacy and control over their data, as it operates without cloud dependencies or telemetry.

These projects showcase the diversity of tools being developed around AI and machine learning frameworks, each catering to specific needs within developer communities ranging from performance optimization to educational resources and practical applications.
Back to all reports