Bookmarks

Top 10 GitHub Repos

hermes-agent

NousResearch's "agent that grows with you" — a long-running, model-agnostic agent harness that accumulates context and skills across sessions rather than starting cold each run. It sits alongside Claude Code, Codex, and OpenClaw as a coding/automation agent and is one of the fastest-accumulating agent repos on GitHub right now. The large open-issue count reflects an unusually active community shaping its direction.

firecrawl

An API to search, scrape, and crawl the web at scale and return LLM-ready markdown — the de facto web-access layer for agents and RAG pipelines. It handles JS-heavy sites, structured extraction, and full-site crawls behind a single call, which is why it shows up as a dependency in a growing share of agent stacks. Momentum remains high as agentic web tasks become a default workload.

browser-use

Makes websites navigable by AI agents — drive real browsers to fill forms, click through flows, and complete tasks online via Playwright under the hood. It's become the go-to library for browser-based automation when an API doesn't exist, pairing well with any LLM. Continued release cadence and broad adoption keep it near the top of agent-automation tooling.

awesome-mcp-servers

The most-referenced catalog of Model Context Protocol servers — a continuously updated index of connectors spanning databases, browsers, cloud services, and developer tools. As MCP becomes the lingua franca for tool access, this list is the practical starting point for anyone wiring servers into Claude, Cursor, or a custom client. The ~1,400 open issues are largely submission/PR traffic, a signal of how fast the ecosystem is still growing.

claude-mem

Persistent memory across sessions for coding agents: it captures what your agent did, compresses it with the model, and re-injects the relevant slice into future sessions. It integrates broadly — Claude Code, OpenClaw, Codex, Gemini, Copilot, OpenCode — using ChromaDB/SQLite under the hood. Memory is one of the hottest agent problems right now, and this is among the most-starred open implementations.

openclaude

Created on 2026-04-01, openclaude has accumulated 28k+ stars in roughly two months with the tagline "runs anywhere, uses anything" — a model-agnostic CLI agent that isn't locked to a single provider. Its fork count (8,600+) is unusually high relative to its age, suggesting heavy hacking and downstream experimentation. Worth watching as a lightweight, portable alternative to vendor-specific agent CLIs.

goose

An open-source, extensible agent (Rust) that goes past code suggestions to actually install, execute, edit, and test against any LLM. It speaks both ACP and MCP, making it a flexible host for the broader tool ecosystem rather than a closed product. Steady commits and a clean extension model keep it a strong pick for teams wanting a self-hostable agent runtime.

terax-ai

A 7MB, terminal-first "AI-native dev workspace" (Tauri + Rust) created on 2026-04-21. It targets developers who want an editor/agent surface that lives in the terminal without the bloat of a full IDE, with cross-platform builds for Linux, macOS, and Windows. The combination of tiny footprint and fast star growth makes it a notable new entrant in the agentic-CLI space.

design-extract

Extracts a website's complete design system in one command — DTCG tokens, semantic/primitive/composite layers — and exposes it as an MCP server for Claude Code, Cursor, and Windsurf. It emits to multiple platforms (SwiftUI, Compose, Flutter, Tailwind v4, Figma variables, shadcn/ui) and includes a CSS health audit and WCAG remediation. Created 2026-04-15, it's a sharp example of a focused, practical MCP tool rather than a general framework.

rmux

A universal Rust multiplexer with a typed SDK to drive any CLI or TUI app from code — native on Linux, macOS, and Windows. Created 2026-05-15, it's effectively a programmable harness for automating terminal programs, which is increasingly relevant for agents that need to operate interactive command-line tools reliably. The youngest repo here, and a quietly important building block for terminal automation.

Honorable Mentions

n8n-io/n8n

Fair-code workflow automation with native AI and 400+ integrations; perennial heavyweight (190k+ stars).

langchain-ai/langchain

The agent engineering platform; still the most common framework dependency for LLM apps.

google-gemini/gemini-cli

Open-source Gemini agent for the terminal, MCP client/server capable.

oritera/Cairn

New general-purpose state-space search agent validated on autonomous penetration testing.

mm7894215/TokenTracker

Local-first token-usage tracker across 22 AI coding tools.