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

agentscope-ai/agentscope

AgentScope 2.0 is a Python (3.11+) agent framework that relies on LLMs' reasoning and tool-use abilities rather than strict prompts or opinionated orchestration. Per its news list, it includes TeamPipeline (a leader agent delegating to member agents), model routing middleware, A2A protocol support, realtime voice agents, and experimental SOP execution.

Developers building multi-agent or voice-enabled agent systems in Python who want a framework with built-in orchestration, routing, and agent-to-agent protocol support should evaluate it.

  • ★ 35,640
  • +★ 480 in 7 days
  • 1.4% growth
  • Apache-2.0
  • README ≈ 6,878 tokens to load into context

AI-written summary

Score90.0
Rank 2

titanwings/distilly

Distilly (formerly Colleague Skill) is a tool that takes source material such as messages, documents, interviews, and public sources and builds a source-grounded Person Profile covering a person's experience, decision patterns, expression, and ways of working. The profile is packaged as an Agent Skill that supported hosts like Claude Code, Codex, OpenClaw, and Hermes can install and invoke.

Developers building agents that need to reproduce a specific person's expertise or communication style, such as a departed teammate's workflow, can look at its three character families (colleague, relationship, celebrity) and its Skill packaging.

  • ★ 26,451
  • +★ 165 in 7 days
  • 0.6% growth
  • MIT
  • README ≈ 7,942 tokens to load into context

AI-written summary

Score80.0
Rank 3

ComposioHQ/composio

Composio is an SDK monorepo (TypeScript, Python, and a CLI) that gives AI agents access to 1000+ pre-authenticated toolkits through per-user sessions. It handles authentication, tool discovery and execution at runtime, and includes provider adapters for frameworks like OpenAI Agents, Anthropic, Claude Agent SDK, and Vercel AI SDK, plus a hosted MCP endpoint per session.

Developers building agents that need to act across many third-party apps without hand-wiring each integration's auth and tool definitions should look at it.

  • ★ 30,988
  • +★ 116 in 7 days
  • 0.4% growth
  • MIT
  • README ≈ 3,833 tokens to load into context

AI-written summary

Score76.1
Rank 4

strands-agents/harness-sdk

Strands Agents is an open-source monorepo with Python and TypeScript SDKs for building AI agents, including a pre-assembled harness, a CLI, and an MCP server for coding agents. It runs in your own process with no hosted control plane and covers lifecycle controls, tools, structured output, MCP, multi-agent patterns, memory, sessions, guardrails, tracing, and evals.

Developers who are about to write their own agent loop in Python or TypeScript should look at it, since it bundles the controls and integrations a hand-rolled loop tends to grow into.

  • ★ 9,080
  • +★ 56 in 7 days
  • 0.6% growth
  • Apache-2.0
  • README ≈ 4,888 tokens to load into context

AI-written summary

Score67.8
Rank 5

Leonxlnx/unlazy

Anti-laziness skill for AI agents. Core: the Depth Tree method, which splits a task N layers deep and gives every leaf the full time budget of the whole task, so effort multiplies with depth. Grounded in 2025-2026 research on model laziness, underthinking and premature completion.

  • ★ 4,341
  • +★ 87 in 7 days
  • 2.0% growth
  • MIT
  • README ≈ 6,427 tokens to load into context

AI-written summary

Score66.1