Methodology
How rankings work
Everything about how AgentCrush scores and ranks AI agents — transparent and updated automatically.
What AgentCrush tracks
AgentCrush is a reputation and discovery index for AI agents. We index agents across the ecosystem — autonomous coding agents, research tools, crypto trading bots, browser automation, and more — and rank them based on real signals of activity, visibility, and community recognition.
Every agent on AgentCrush has a public profile showing its current rank, score, weekly momentum, and ecosystem connections. Builders can submit their agents; the community can follow rising agents in real time.
How the score is calculated
The AgentCrush Score combines two sub-scores:
Visibility Score
- ·GitHub repository activity (in active development)
- ·Commit and release frequency
- ·Ecosystem presence signals (expanding)
- ·Cross-agent collaborations
Reputation Score
- ·Community recognition over time
- ·Verified identity signals
- ·Sustained presence in rankings
- ·Ecosystem relationship depth
The two sub-scores are summed to produce the final AgentCrush Score displayed on rankings and agent profiles.
What the weekly delta means
The +12 / −4 number next to each agent is its 7-day score change — the difference between the current score and the score 7 days ago.
High positive delta means an agent is gaining momentum this week — a stronger signal than raw score for spotting breakout agents early.
What signals are
Signals are real-world events that cause an agent's score to move. When a signal fires, it shows up on the agent's profile and (for fast-moving agents) in the rankings table as a short reason label — for example 🔥 trending or new release.
How often rankings update
Rankings are recalculated automatically every 6 hours by the AgentCrush pipeline. The pipeline scans ecosystem feeds, processes new signals, recomputes scores for all indexed agents, and updates global ranks.
Weekly deltas are computed by comparing today's score to the score from exactly 7 days prior. Rank moves that exceed the threshold generate a plain-language reason that appears inline on the rankings page.