Top Picks

Smriti is an open-source system for shared reasoning state across coding agents like Claude Code and Codex. It treats project reasoning like version-controlled state, with local setup, a CLI, backend, and frontend for coordinating multiple agents on the same repo.

Open-source AI SEO optimizer CLI that audits pages, generates meta tags, analyzes keyword gaps, and drafts blog outlines. It can run free for audits and uses a KianBot API key for AI-assisted features.

An AI email assistant that lets people send a question or long email to [email protected] and get back a summary, answer, actions, dates, or draft response by email, with no signup and a free tier.

An agent skill for starting or extending Django projects, with ready-made scaffolding, stack choices, and safeguards for common production pieces like auth, jobs, Redis, email, and deploy setup.

A WordPress weather widget plugin with shortcode, Gutenberg block, sidebar support, auto-location, multilingual output, and a lightweight embedded UI.

A free, private hormone interpretation tool that calculates free testosterone and maps testosterone-related markers into a feedback-pathway view with lab-specific guidance.

A Claude plugin that adds interactive architecture diagrams to Plan Mode, letting users review, annotate, and revise system designs in real time.

An open-source research framework for structured LLM evaluation, claim verification, and source-grounded reflective reasoning. The repo describes modular components for retrieval, semantic scoring, skeptical claim checking, and benchmark-style epistemic assessment.

A Rust-based MCP server that lets AI agents chat with each other, manage channels, and be monitored through a web page.

Lumabri is an open-source distributed inference tool that runs large AI models across multiple computers and servers. It includes a terminal UI and browser interface, with setup, networking, memory budgeting, and calibration details documented in the README.

Anthropic’s research post shows Claude proving security properties across 123 open-source libraries and discusses how LLMs can support formal verification workflows for real software.

Self-hosted memory layer for coding agents. It stores notes as Markdown in a local folder and exposes them over MCP so tools like Claude Code, Codex, and Cursor can recall shared context.

A Claude Code plugin marketplace with 42 skills for building, shipping, and announcing apps across App Store, Google Play, Mac, Windows, and web workflows.

Hybro is an open-source interoperability layer for AI agents, with a FastAPI backend, Next.js dashboard, CLI setup, and support for the A2A protocol and MCP server integration.

A Claude Code plugin that wraps MCP tool use in a single sandboxed JavaScript program, reducing tool-call chatter and keeping intermediate results out of context.

Galahad is an AI memory layer that saves and restores model KV cache, supports document recall, and plugs into vLLM, SGLang, and llama.cpp. The page also shows agent features like replay, branching, rewind, and fault-finding.

A browser-game playing agent that uses an LLM to train per-game rules and then makes live decisions through either a hosted engine (Jev) or a local engine (Laya).

An experimental repository showing an LLM-generated binary workflow: natural-language specs are turned into raw executable bytes, then tested in a loop with examples like a hello world, HTTP server, and Tetris.

A Vercel tutorial showing how to wire Jev into a TanStack Start app using TanStack AI and the Vercel Gateway adapter, with typed decision routing and server-side validation examples.

FDA is a developer tool for giving AI agents temporary, session-scoped access to machines with audit logs instead of standing SSH access.

Open-source Claude Skill plus MCP integration for job search and interview prep, with live job data, role-specific interview intel, answer scoring, resume scoring, and voice mock interviews as an upgrade path.

A local coding-agent setup that turns Ollama models into a private, verified team of specialist agents for coding workflows. It includes hardware-aware model selection, role assignment, permission-restricted workflows, and validation probes.

AgentTrace is a Python-based observability and runtime self-healing system for AI agent workflows. It logs multi-step tool calls, surfaces latency issues, and attempts to repair malformed LLM tool arguments before they crash a run.

An interactive Pong demo that compares Jev against Gemini 3.8 Flash, Claude Haiku 4.5, and GPT-5.6 Sol, highlighting response-time differences in a real-time game loop.