AI agent development has exploded in 2026. The challenge isn't finding tools — it's knowing which ones exist and how they compare.
I've been building AgDex.ai — a curated directory of 370+ AI agent tools, available in 4 languages (EN/ES/DE/JA).
What's Covered
🏗️ Core Frameworks
The foundational orchestration layers for building agents:
- LangChain / LangGraph — the most adopted, graph-based workflows
- CrewAI — role-based multi-agent teams
- AutoGen (Microsoft) — conversational multi-agent framework
- OpenAI Agents SDK — native tool use + handoff patterns
- Google ADK — Google's agent development kit
- Mastra — TypeScript-native agent framework
☁️ Cloud & Inference
Where your agents actually run:
- Together AI — fast open-source model inference
- Replicate — 100k+ models via API
- Fireworks AI — low-latency fine-tuned model serving
- Vast.ai / Lambda Labs — cheap GPU rentals
- Modal / Beam Cloud — serverless GPU compute
🔍 Observability & Evaluation
You can't improve what you can't measure:
- Langfuse — open-source LLM tracing (self-hostable)
- LangSmith — LangChain's native observability
- Helicone — lightweight proxy with analytics
- Arize Phoenix — ML observability platform
- Ragas / DeepEval — automated RAG + agent evaluation
🌐 Browser & Web Agents
The fastest growing category in 2026:
- Browser Use — Python library for LLM browser control
- Stagehand — Playwright + LLM, type-safe
- Skyvern — vision-based browser automation
- Firecrawl — structured web scraping for LLMs
- Tavily / Exa — search APIs built for agents
🧠 Memory & Knowledge
Giving agents persistent, long-term context:
- Mem0 — memory layer for personalized AI (formerly Memory.ai)
- Zep — temporal knowledge graphs for agents
- Letta (formerly MemGPT) — OS-inspired memory management
- Graphiti — episodic memory via knowledge graphs
🔀 LLM Gateways
Route, fallback, and manage LLM costs:
- LiteLLM — 100+ models, self-hostable proxy
- OpenRouter — 300+ models, generous free tier
- Portkey — production observability + prompt management
🛡️ Safety & Guardrails
For responsible agent deployment:
- NeMo Guardrails (NVIDIA) — programmable safety rails
- LLM Guard — input/output scanning
- Presidio — PII detection and anonymization
Why a Directory?
When I started building with agents, I kept discovering great tools through random blog posts and GitHub stars. There was no single place to:
- Compare tools by category
- Find the international equivalents (many great JP/DE/ES tools)
- Understand pricing models at a glance
So I built it.
Numbers
- 370+ tools indexed (growing weekly)
- 7 categories: frameworks, cloud, tools, ecosystem, LLM, community, tutorials
- 4 languages: all names and descriptions in EN/ES/DE/JA
- Fully open: the data is on GitHub
What's Missing?
The directory is far from complete. I'm actively looking for:
- Regional tools dominant in specific markets (JP, DE, FR, KR)
- Niche evaluation frameworks
- New MCP servers and integrations
- Enterprise-specific tooling
If you know a tool that should be listed: the GitHub repo has an issues page, or reach me at agdex.ai@gmail.com.
Built with: vanilla HTML + TailwindCSS + JSON data, hosted on Cloudflare Pages. Zero build step, maximum speed.
Top comments (0)