Forkcast Weekly #12: agency-agents, code-review-graph, open-code-review, jcode, i-have-adhd
This episode covers 30 trending open-source projects from this week including a complete AI agency with 137k stars, a code intelligence graph that reduces
Watch Episode
Episode Summary#
This episode covers 30 trending open-source projects from this week including a complete AI agency with 137k stars, a code intelligence graph that reduces AI token waste, Alibaba battle-tested code review tool, a Claude Code ADHD-friendly skill, persistent AI memory with knowledge graphs, a financial markets foundation model with 34k stars, an open-source voice studio, and the 261k-star superpowers framework. Also featured are croc file sharing, ktransformers for heterogeneous LLM inference, likec4 live architecture diagrams, and the original Apollo 11 guidance computer source code.
Repositories Covered#
agency-agents#
Building an AI-powered agency requires assembling dozens of specialized roles from scratch — frontend, community, content, QA, each needing their own prompts and workflows. 136.7k stars, 22.3k forks. Each agent is a specialized expert with personality, processes, and proven deliverables. Covers frontend, Reddit, whimsy injection, reality checking, and more.
code-review-graph#
AI coding tools waste tokens reading entire files when they only need context from a few related functions. Large-repo workflows burn through context windows fast. 26.4k stars. Persistent knowledge graph of your codebase. Benchmarked context reductions on reviews. Works with MCP and CLI. Local-first, no cloud required.
open-code-review#
Manual code review doesn’t scale. AI-only reviews miss domain-specific bugs like NPE, thread-safety, and injection vulnerabilities that static analysis catches. 13.1k stars, 894 forks. Built-in fine-tuned ruleset covering NPE, thread-safety, XSS, SQL injection. OpenAI and Anthropic compatible. Go backend, TypeScript frontend.
jcode#
Existing coding agent harnesses lack taxonomy-aware discovery and struggle with complex multi-file refactoring tasks. 11.5k stars, 6,011 commits. MIT licensed. Financial-data category taxonomy recently added. Active development with 42 branches.
i-have-adhd#
AI coding agents bury the answer under walls of explanation. You ask a simple question and get five paragraphs before the one-liner you needed. 10.2k stars. Single skill file that changes agent output format. Works with Claude Code, Copilot, and other agents. MIT licensed.
cognee#
AI agents forget everything between sessions. You rebuild context from scratch every time, wasting tokens and losing critical decisions. 29.4k stars, 205+ contributors. Self-hosted knowledge graph engine. Python 85.6%. Persistent memory across sessions.
Kronos#
General-purpose language models don’t understand financial time series, market microstructure, or the unique vocabulary of trading. 33.9k stars, 5.7k forks. MIT licensed. Foundation model specifically trained on financial market data and language.
text-to-cad#
CAD and hardware design tools have steep learning curves. Translating natural language descriptions into parametric 3D models requires expert knowledge. 10.4k stars, 314 commits. MIT licensed. Agent skills for CAD, robotics, and hardware design. 84 branches, active development.
ai-agent-book#
AI agent development is fragmented across blog posts and tutorials. There’s no single authoritative resource covering design principles and production engineering. 20k stars, 2k forks. Full book text, compiled PDF, and per-chapter code examples. Apache-2.0 licensed.
voicebox#
AI voice tools are either cloud-locked, expensive, or lack the studio-quality controls professionals need for voice cloning and creation. 46.8k stars, 5.7k forks. Open-source voice studio with clone, dictate, and create capabilities. 66 branches, active development.
awesome-claude-skills#
Claude’s skill ecosystem is growing fast but there’s no central index. Developers waste time searching for the right skills and resources. 70.7k stars, 7.9k forks. Curated list covering Claude Skills, resources, and workflow customization tools.
Pumpkin#
Java Minecraft servers are memory hogs. Running a vanilla server for a handful of friends shouldn’t require a dedicated machine. 9.8k stars, 660 forks. Written in Rust for performance. 2,495 commits. Self-hosted Minecraft server alternative.
worldmonitor#
Monitoring global events across dozens of sources requires manual aggregation. No single dashboard combines news, geopolitics, and infrastructure tracking. 74.3k stars, 11.2k forks. Go backend, 5,021 commits, 280 branches. AI-powered news aggregation and situational awareness.
croc#
Sending files between devices is needlessly complex. AirDrop is Apple-only, email has size limits, and cloud services require accounts. 38.5k stars, 1.5k forks. End-to-end encrypted. Works across any OS. No accounts, no cloud. MIT licensed.
ktransformers#
Running large LLMs requires expensive GPU clusters. CPU-GPU heterogeneous inference could dramatically reduce hardware costs. 19k stars, 1.5k forks. Heterogeneous CPU-GPU inference framework. 56 branches, active development.
likec4#
Architecture diagrams go stale the moment code changes. Manual diagramming can’t keep pace with fast-moving codebases. 5.2k stars, 342 forks. Live diagrams generated from code. TypeScript. 191 releases.
harper#
Grammar checkers send your text to the cloud. Privacy-conscious writers and teams need an offline alternative that’s fast enough for real-time use. 13.5k stars, 513 forks. Rust-powered, offline-first. 4,460 commits, 226 releases. Automattic-backed.
OmniRoute#
AI coding tools burn through paid API credits fast. Rate limits, lockouts, and bills pile up from tool-call overhead alone. 30.3k stars, 3.9k forks. 290+ providers, 90+ free. 500+ models. RTK+Caveman compression saves 15-95% tokens. Desktop/PWA. MIT licensed.
ego-lite#
AI agents can’t use your logged-in browser sessions. They start fresh every time, requiring re-authentication and losing cookies. 3.8k stars, 189 forks. Zero cost, zero config. Shares logged-in browser state with Codex, Claude Code. 237 commits.
lingbot-map#
3D scene reconstruction from video streams is slow and requires per-scene optimization. Real-time 3D maps from live camera feeds weren’t possible. 15.4k stars, 1.6k forks. Feed-forward architecture. Streaming data input. Apache-2.0 licensed. 110 commits.
ai-engineering-from-scratch#
AI engineering tutorials are scattered, outdated, or too academic. You need a practical roadmap that takes you from zero to building and shipping. 43.6k stars, 7.3k forks. Python 66.1%. Practical, project-based curriculum. Covers building and shipping AI products.
claude-cookbooks#
Claude’s capabilities are vast but underexplored. Developers need practical examples and recipes to unlock advanced use cases. 50k stars, 5.9k forks. Official Anthropic repository. 146 branches. Notebooks and recipes for RAG, agents, and more.
AstrBot#
Building multi-platform AI chatbots requires integrating each messaging service separately. No unified framework handles IM platforms, LLMs, and plugins together. 38.1k stars, 2.7k forks. 176 branches, 255 releases. Integrates multiple IM platforms with LLMs and plugin system.
superpowers#
AI coding agents are unpredictable. They write tests that don’t test anything, refactor without understanding architecture, and ship broken code. 261k stars, 23.3k forks. Enforces TDD, spec-first workflows, git worktree isolation, two-stage code review. v6.2.0 with SDD plan-scoped workspace.
Apollo-11#
Understanding how the Apollo missions actually worked at the code level requires digging through NASA archives that aren’t easily accessible. 71.5k stars, 7.9k forks. Original AGC source code. 588 commits. Assembly language. Historical software preservation.
open-seo#
SEO tools like Semrush and Ahrefs cost hundreds per month. Small teams and indie developers can’t afford enterprise-grade keyword research. 8k stars, 878 forks. TypeScript 97.2%. 402 commits, 31 releases. Self-hosted SEO analysis platform.
superfile#
Terminal file managers are either ugly, slow, or lack the features of GUI file managers. Developers want a beautiful, responsive TUI file manager. 19.9k stars, 630 forks. Go-based. MIT licensed. Modern terminal UI with file management features.
openship#
Deployment platforms like Vercel and Railway lock you into their infrastructure. Self-hosted alternatives are complex to set up and maintain. 8.6k stars, 682 forks. 13 releases, 2 branches. Self-hosted deployment platform with simple setup.
RuView#
Spatial awareness and vital sign monitoring currently require cameras or wearable sensors. WiFi signals could provide this data passively. 86.5k stars, 11.5k forks. Rust 56.6%. WiFi-based presence detection, vital signs monitoring. 1,153 commits.
buzz#
Team communication tools are siloed. No platform enables truly decentralized, peer-to-peer group communication with AI agent integration. 12.2k stars, 980 forks. Rust 47.8%, TypeScript 35%. 1,857 commits, 461 branches. Hive mind communication architecture.
Watch#
- Video: https://www.youtube.com/watch?v=YR9SzaJ5c80
- Cover: https://i.ytimg.com/vi/YR9SzaJ5c80/hqdefault.jpg
Notes#
Transcript and notes will be added from Forkcast output artifacts.