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

upstash/context7

Context7 is a platform that pulls up-to-date, version-specific library documentation and code examples from the source and supplies them to coding agents. It is distributed via an MCP package (@upstash/context7-mcp) and integrations for Claude, ChatGPT, Cursor, and VS Code, and is invoked by adding "use context7" to a prompt.

Developers whose coding agents produce outdated examples, nonexistent APIs, or answers for old package versions can use this to ground the agent in current docs.

  • ★ 63,741
  • +★ 405 in 7 days
  • 0.6% growth
  • MIT
  • README ≈ 6,979 tokens to load into context

AI-written summary

Score91.4
Rank 2

1Panel-dev/MaxKB

MaxKB is an open-source platform for building enterprise-grade agents, combining a RAG pipeline (document upload or crawling, automatic splitting and vectorization), a workflow engine with a function library, and MCP tool use. It works with both private and public LLMs, handles text, image, audio and video, and runs via Docker on a Django/Vue.js stack with PostgreSQL + pgvector.

Developers who want a self-hosted, GPL-licensed platform for knowledge-base Q&A and workflow-driven agents that can be embedded into existing business systems without coding should look at it.

  • ★ 23,048
  • +★ 48 in 7 days
  • 0.2% growth
  • GPL-3.0
  • README ≈ 2,007 tokens to load into context

AI-written summary

Score77.1
Rank 3

Fosowl/agenticSeek

AgenticSeek is a local alternative to Manus AI that runs on your own hardware with local reasoning models (e.g., via Ollama or LM Studio). It uses multiple agents to browse the web, write and run code in languages like Python, C, Go, and Java, and plan multi-step tasks, with voice input listed as in progress.

Developers who want an autonomous browsing and coding agent that keeps data on their own machine, rather than depending on cloud APIs, may want to look at its multi-agent setup, which uses SearxNG and Redis via Docker.

  • ★ 27,551
  • +★ 37 in 7 days
  • 0.1% growth
  • GPL-3.0
  • README ≈ 17,018 tokens to load into context

AI-written summary

Score47.1
Rank 4

HKUDS/DeepCode

DeepCode is an open-source agentic coding project from HKUDS that describes itself as an agent harness with loop engineering and multi-agent orchestration. It offers TUI, Desktop, and Web clients (`deepcode`, `deepcode desktop`, `deepcode web`) over one shared local service that handles Projects, Sessions, models, Skills, permissions, Goals, and Automations.

Developers building or evaluating coding agents may want to look at it for its multi-agent harness design and shared local service, which also has an associated arXiv paper.

  • ★ 16,757
  • +★ 22 in 7 days
  • 0.1% growth
  • MIT

AI-written summary

Score47.1
Rank 5

gptme/gptme

gptme is an open-source, provider-agnostic agent that runs in a terminal and ships with local tools including shell, Python, web browsing, and vision. It supports Anthropic, OpenAI, Google, OpenRouter, and local models via Ollama or OpenAI-compatible servers, and can serve as a base for building persistent autonomous agents.

Developers who want a terminal-based, local-first agent they can extend with plugins, skills, MCP/ACP integrations, or turn into an autonomous agent should look at it.

  • ★ 4,466
  • +★ 9 in 7 days
  • 0.2% growth
  • MIT
  • README ≈ 15,766 tokens to load into context

AI-written summary

Score45.7