The live index of every public AI agent.
Find and compare the best AI agents and foundation models as they emerge. Scored by an open formula. Free to read, no login needed.
The 100 best AI agents, ranked.
Top 100 application agents by AgentScore. Auto-rebalanced weekly. Sortable by movement, filterable by capability.
Open the index →
The 50 best foundation models.
Claude, GPT, Llama, Gemini, Mistral, DeepSeek and more, tracked as their own asset class. Compare GPT vs Claude directly with an overlaid chart.
Open the model board →
What is AgentTape?
AgentTape is a live, autonomously-populated index of every public AI agent. Each agent is its own stock ticker with a transparent 0-100 AgentScore, four pillar scores (Adoption, Quality, Momentum, Community), and a chart of how it has moved over time. Click any agent to see why its score is what it is. Every input is published.
We do not curate. The discovery service watches GitHub, Hugging Face, MCP registries, npm and PyPI, arXiv, and Hacker News, plus the OpenRouter model catalogue, then admits agents on its own schedule. If a project starts gaining real traction, it shows up here without anyone asking.
What can I use this for?
Pick the best AI agent for my use case
Filter by capability (code generation, browsing, research, RAG, multi-agent) and see the top stocks ranked by AgentScore. Then drop the top 2-4 onto the comparison page and decide.
Compare two foundation models
Stick Claude and GPT side-by-side on the compare page. You get an overlay chart of their AgentScore over 30 days, pillar bars next to each other, plus a clean 'best at X' verdict per model.
See which agents are gaining traction
Trending shows the biggest score moves over 1h / 24h / 7d / 30d, split by application agents vs foundation models. Rank-movement arrows show who climbed and how far.
Track newly-released agents and models
The discovery feed surfaces new admissions in real time, the day they're noticed by GitHub, npm, Hugging Face or OpenRouter. Catch the next AutoGPT before everyone else has heard of it.
Today's top stocks
Live snapshot. Refreshes whenever the page is recrawled.
How the AgentScore works
A 0-100 headline backed by four pillars. The weights below are the defaults; quality's 30% is redistributed pro-rata when an agent is unrated so a young project isn't penalized for the absence of benchmark coverage.
- Adoption (35%): z-score across stars, downloads, package counts, registry presence
- Quality (30%): mean z-score across benchmark results. Unrated when no benchmark exists, never zero
- Momentum (20%): 7-day and 30-day rate-of-change blend on adoption signals
- Community (15%): contributors, HN points, Reddit points
Manipulation rules (star spike without contributor diversity, HF download surge without matching GitHub activity, coordinated HN posting) exclude implicated signals from that day's score. Read the full methodology.
Browse by category
AgentTape is open source and built in public. Read the methodology for the math, or the repo for the source. No login. No paywall. No newsletter wall.