Everyone lists the trending repos.
We tell you which ones to use.

firecrawl

★ 174k

firecrawl

Turns any website into clean markdown an LLM can read, including pages that need JavaScript to render.

The catch: AGPL-3.0. Fine to use, but if you self-host it inside a product you ship to customers, you may owe them your source. Their cloud avoids the question.

Scrapling

★ 77k

D4Vinci

A Python scraping library that adapts when a site changes its markup, instead of breaking silently.

The catch: It is a young project carried by one main maintainer. The permissive BSD licence limits the damage, but plan for the bus factor.

cc-switch

★ 130k

farion1231

AI-powered desktop assistant

The catch: dependency on multiple AI services

Agent-Reach

★ 77k

Panniantong

CLI tool for AI agents to search internet platforms

The catch: Dependence on unmaintained platform APIs or CLI complexity

headroom

★ 68k

headroomlabs-ai

Pre-LLM token compressor for logs, JSON, and tool outputs, saving 20-95% tokens.

The catch: Compression may lose nuance in critical logs; needs careful testing per use case.

ponytail

★ 116k

DietrichGebert

AI code review and suggestion tool

The catch: may introduce AI bias and require significant training data

agentscope

★ 30k

agentscope-ai

Open-source agent and automation framework

The catch: Steep learning curve and potential lock-in to custom architecture

browser-use

★ 112k

browser-use

Lets an LLM drive a real browser — click, type, scroll — so it can use websites that have no API.

The catch: Every run costs model tokens and it fails in ways a scripted scraper does not. Treat it as best-effort, never as a payments-critical path.

open-design

★ 86k

nexu-io

Local-first design tool with AI coding agent

The catch: Steep learning curve and potential AI model lock-in

n8n

★ 203k

n8n-io

A drag-and-drop workflow builder that wires your apps together and can call an LLM at any step.

The catch: The licence is source-available, not open source — you can self-host and use it internally, but you cannot resell it as your own hosted product.

open-webui

★ 150k

open-webui

A polished self-hosted chat interface over your own models or API keys, with document upload and user accounts built in.

The catch: The licence now restricts removing its branding above a user threshold, so read it before white-labelling. You also own the hosting.

ECC

★ 244k

affaan-m

AI agent performance optimization system

The catch: immaturity and limited documentation

deer-flow

★ 81k

bytedance

AI agent framework for automation

The catch: complexity and steep learning curve

dify

★ 154k

langgenius

A visual builder for LLM apps: prompt chains, retrieval over your documents, and agent steps without writing the plumbing.

The catch: Source-available licence with restrictions on multi-tenant resale, and the visual layer becomes a ceiling once your logic gets specific.

harness-sdk

★ 7.0k

strands-agents

SDK for building AI agents in Python & TypeScript

The catch: High ops burden and potential lock-in to custom agent framework

GenericAgent

★ 14k

lsdefine

Python agent for automation and system control

The catch: Steep learning curve and potential system instability

MaiBot

★ 5.9k

Mai-with-u

LLM-based chatbot agent in Python

The catch: GPL-3.0 license may restrict commercial use

lobehub

★ 82k

lobehub

A self-hostable chat interface that puts several model providers and a library of configured agents behind one login.

The catch: You take on hosting, auth and the API bill, and the pace of change means upgrades are not free. Source-available licence, not fully open.

ragflow

★ 90k

infiniflow

A document-understanding engine that parses messy PDFs and scans properly before answering questions about them.

The catch: It is a heavyweight system to run and the open-issue count is large. Budget real infrastructure time, not an afternoon.

mem0

★ 64k

mem0ai

A memory layer that lets an AI assistant remember facts about a user across separate conversations.

The catch: Memory systems fail in embarrassing ways: recalling the wrong fact is worse than recalling nothing. It also becomes personal data you now store.

anything-llm

★ 65k

Mintplex-Labs

A single application that turns a folder of your documents into a chat assistant, as a desktop app or a hosted install.

The catch: Retrieval quality is only as good as your documents. Messy, contradictory files produce confident, wrong answers.

langchain

★ 145k

langchain-ai

The most widely used Python toolkit for chaining model calls, tools and retrieval into an application.

The catch: The abstractions move fast and break, and they hide behaviour you will eventually have to debug. Many teams end up unwinding them.

PaddleOCR

★ 88k

PaddlePaddle

Reads text out of scans, photos and PDFs, including tables and forms, and returns structured output.

The catch: The documentation and issue tracker lean heavily Chinese, so debugging takes longer if you do not read it. Accuracy varies with scan quality.

cognee

★ 30k

topoteretes

Self-hosted knowledge graph engine for AI agents

The catch: High ops burden and potential for knowledge graph complexity

litellm

★ 57k

BerriAI

One API in front of every model provider, with per-key spend limits, logging and automatic failover.

The catch: The open-issue count is very large and the surface area is wide; pin your version and read release notes before upgrading.

langfuse

★ 34k

langfuse

AI engineering platform for LLMs

The catch: unclear custom license may limit commercial use

promptfoo

★ 25k

promptfoo

AI model testing and evaluation framework

The catch: steep learning curve and dependency on maintainers

gitleaks

★ 29k

gitleaks

Secrets scanner for Git repos

The catch: False positives and noise in output

opik

★ 22k

comet-ml

Open-source LLM evaluation and monitoring tool

The catch: Steep learning curve and potential ops burden

semantic-router

★ 5.3k

vllm-project

Mixture-of-Models router for LLM inference

The catch: Immature with 333 open issues and limited documentation

phoenix

★ 11k

Arize-ai

AI model monitoring and evaluation tool

The catch: unclear custom license may limit commercial use

flashinfer

★ 6.3k

flashinfer-ai

LLM serving kernel library

The catch: high ops burden and CUDA dependency

openevolve

★ 7.3k

algorithmicsuperintelligence

Open-source AlphaEvolve implementation

The catch: Immaturity and limited community support

rtk

★ 78k

rtk-ai

CLI proxy reducing LLM token usage

The catch: dependency on LLM compatibility

DeepSeek-Reasonix

★ 35k

esengine

AI coding agent for terminal

The catch: dependency on unstable AI models

code-review-graph

★ 31k

tirth8205

Local code intelligence graph for AI coding tools

The catch: Steep learning curve and potential performance overhead

codeburn

★ 9.7k

getagentseal

Local tool to track AI coding costs

The catch: Dependence on a single maintainer for updates

OpenHands

★ 85k

OpenHands

An AI developer you hand a task to — it plans, writes code, runs it, and iterates until the task passes.

The catch: Model costs add up fast on long tasks, and unreviewed output is a liability. It needs a human gate, which eats much of the saving.

loop-engineering

★ 11k

cobusgreyling

AI coding agent tools and patterns

The catch: immature and limited community support

serena

★ 29k

oraios

AI-powered coding toolkit

The catch: Immature and limited community support

agentic-awesome-skills

★ 46k

sickn33

Local agent control plane for skill discovery

The catch: Complexity and steep learning curve

yazi

★ 42k

sxyazi

Fast terminal file manager

The catch: Steep learning curve for non-Rust users

ruflo

★ 68k

ruvnet

Agent meta-harness for AI workflows

The catch: Steep learning curve and immature ecosystem

superpowers-zh

★ 7.9k

jnMetaCode

AI coding tool with Chinese support

The catch: Niche focus and limited community support outside China

qwen-code

★ 27k

QwenLM

AI coding agent in terminal

The catch: immaturity and potential ops burden

bruno

★ 47k

usebruno

An API client that stores collections as plain files in your repo, so they version and review like code.

The catch: The open-issue count is high and the desktop app is less polished than the incumbent. Some advanced features sit behind a paid tier.

supervision

★ 50k

roboflow

The connective tissue around vision models: drawing boxes, counting objects, tracking across frames, zone logic.

The catch: It is a helper library, not a model. It solves the last mile and nothing before it.

screenpipe

★ 21k

screenpipe

Screen recording and AI agent integration tool

The catch: Unclear custom license may limit commercial use

ultralytics

★ 61k

ultralytics

Ready-to-run object detection and segmentation models you can train on your own images in an afternoon.

The catch: AGPL-3.0. If you ship this inside a commercial product without opening your source, you need their paid licence. Price that before you build on it.

MoneyPrinterTurbo

★ 118k

harry0703

AI-powered short video generator

The catch: dependency on unstable AI models and ffmpeg

unsloth

★ 75k

unslothai

Makes fine-tuning open models substantially faster and small enough to fit on one consumer GPU.

The catch: It is excellent at a job most small teams should not be doing yet. The prerequisite is a labelled dataset you do not have.

OpenMontage

★ 53k

calesthio

Open-source video production system with AI tools

The catch: AGPL-3.0 license may restrict commercial use

VoxCPM

★ 36k

OpenBMB

Open-source text-to-speech synthesis

The catch: High computational requirements and limited support

index-tts

★ 24k

index-tts

Open-source text-to-speech system

The catch: Unclear custom license may limit commercial use

sherpa-onnx

★ 14k

k2-fsa

Offline speech-to-text and text-to-speech engine

The catch: High maintenance and customization required for specific use cases

voice-pro

★ 13k

abus-aikorea

Gradio WebUI for TTS and voice cloning

The catch: GPL-3.0 license may limit commercial use

SkillSpector

★ 15k

NVIDIA

AI security scanner for agent skills

The catch: May require significant customization and integration effort

shannon

★ 47k

KeygraphHQ

An autonomous agent that tests your own web application for security holes and writes up what it finds.

The catch: Only ever point it at systems you own and are authorised to test. It supplements a real pentest; it does not satisfy an auditor who wants one.

pentagi

★ 22k

vxcontrol

AI-powered penetration testing tool

The catch: immaturity and potential false positives

trivy

★ 38k

aquasecurity

Open‑source scanner for container images, IaC, and code secrets

The catch: Requires regular DB updates, can generate false positives, and lacks commercial support if you hit edge cases