Pithelo

Five agent repos gaining momentum. Every one measured.

Ranked by stars gained over the last 7 days, growth rate and recent commit activity.

Rank 1

thedotmack/claude-mem

claude-mem is a persistent memory system that captures what an agent does during sessions, compresses it with AI into semantic summaries, and injects relevant context into later sessions. It is described as working with Claude Code, OpenClaw, Codex, Gemini, Hermes, Copilot, OpenCode and others, and installs via `npx claude-mem install --ide <name>`.

Developers whose agents lose project context between sessions or reconnects can look at this as a ready-made way to carry that context forward.

  • ★ 110,982
  • +★ 7,495 in 7 days
  • 7.2% growth
  • Apache-2.0
  • README ≈ 8,459 tokens to load into context

AI-written summary

Score97.5
Rank 2

ruvnet/ruflo

Ruflo is an agent meta-harness for Claude Code and Codex that wraps them with 100+ specialized agents, coordinated swarms, persistent memory, federation across machines, and security guardrails. It installs as a Claude Code plugin or via npx, exposes an MCP server for other clients, and includes a console and an optional terminal UI.

Developers building multi-agent workflows on Claude Code or Codex who want swarm coordination, cross-session memory, and an MCP-accessible toolset in one package should look at it.

  • ★ 80,683
  • +★ 1,086 in 7 days
  • 1.4% growth
  • MIT
  • README ≈ 14,056 tokens to load into context

AI-written summary

Score87.5
Rank 3

feder-cr/git-ai

Git designed for AI agents.

  • ★ 1,043
  • +★ 41 in 7 days
  • 4.1% growth
  • No licence declared
  • README ≈ 3,734 tokens to load into context

AI-written summary

Score72.5
Rank 4

browser-use/browser-use

browser-use is an open-source Python library for building AI agents that control a web browser, usable with any LLM and a local or cloud browser. The same project also offers Browser Use Cloud, a hosted service with stealth browsers and an agent API, plus a CLI that registers a skill for coding agents.

Developers building agents that need to navigate and act on websites can choose between running the library locally or using the hosted browsers and agent API.

  • ★ 119,890
  • +★ 652 in 7 days
  • 0.5% growth
  • MIT
  • README ≈ 7,402 tokens to load into context

AI-written summary

Score69.4
Rank 5

AMAP-ML/LongHorizon-Harness

LongHorizon-Harness is a Python harness that wraps existing coding agents (Claude Code, Codex, OpenCode, DeepSeek Harness) in a plan–act–verify–checkpoint loop so they can run across desktop apps and the terminal for extended periods. Each round runs in a fresh context against saved, verified state, with independent auditing and recovery from failures or context refreshes.

Developers running agents on long, multi-step computer-use tasks who need durable state, verification of real results, and resumable progress without training a new model should look at it.

  • ★ 1,833
  • +★ 51 in 7 days
  • 2.9% growth
  • MIT
  • README ≈ 13,186 tokens to load into context

AI-written summary

Score69.4