AgentCrush/Dataset
For researchers · journalists · data teams
The AgentCrush Dataset
Daily snapshots of 1,400+ AI agents tracked across the ecosystem. Liveness-scored, methodology-versioned, and anchored on-chain every night. All endpoints are free and open — no account required for research use.
What the dataset is
GitHub, Virtuals, Agentverse, ERC-8004, and direct submissions — cross-protocol.
Multi-signal corroboration: stars, citations, deployments, protocol presence, on-chain registrations.
Daily scoring run since April 2026. Liveness, tier, signal breakdown per agent per day.
SHA-256 Merkle root of the daily snapshot committed to Base (public tx). Methodology-versioned.
Daily liveness index published free. History included in the API response.
Use commercially, republish, embed — with attribution and a link.
How to access it
All endpoints are CORS-open, no API key required for the free tier. Paid endpoints ($0.05) accept x402 payment or a Pro key. Full OpenAPI spec at /api/openapi.json.
/api/ghost-index/v1Daily liveness index — full history, category breakdown, delta
/api/agent-economy/llm-summaryMachine-readable ecosystem metrics: indexed count, snapshot total, evidence-ranked count
/api/trust/evaluate?handle=<handle>Trust evaluation for a single agent — tier, liveness, reason codes
/api/agent/<handle>/llm-summaryAgent-level summary for LLM consumption — profile, signals, ranking position
/api/oracle/attest?metric=liveness&handle=<handle>Ed25519-signed, timestamped liveness attestation for a single agent
/api/oracle/attest?metric=ghost_indexSigned attestation of the current Ghost Index liveness score
/api/proof-of-index/v1Latest on-chain anchor — Merkle root, Base tx hash, snapshot date
/api/agents/find?q=<capability>Discovery — top 3 matching agents ranked by trust + liveness
/api/agents/find/full?q=<capability>Full-list discovery with payment rails and endpoints
Provenance and auditability
Every ranking score is produced by a published Postgres view. The nightly snapshot commits a SHA-256 Merkle root of all daily scores to a public Base transaction — the on-chain hash is the receipts layer for any later audit.
Methodology is versioned and stable enough to cite. If scoring rules change, a new version string is issued — old attestations remain interpretable against the version that produced them.
BibTeX
@misc{agentcrush2026,
title = {{AgentCrush: The Evidence-Ranked Index of the AI Agent Economy}},
author = {{Patyi, Kristof and the AgentCrush team}},
year = {2026},
url = {https://agentcrush.xyz},
note = {Live methodology at https://agentcrush.xyz/methodology.
Daily snapshots CC-BY-4.0. Dataset page:
https://agentcrush.xyz/dataset.},
}Suggested inline citation
Article / blog post
“According to the AgentCrush Ghost Index, 55% of the 1,428 AI agents it tracks showed observable activity in the last 30 days (agentcrush.xyz/ghost-index, accessed YYYY-MM-DD).”
Academic paper (abbreviated)
“We use the AgentCrush index [1] to measure agent liveness across 1,428 tracked agents. Daily snapshots are anchored on-chain and available under CC-BY-4.0 atagentcrush.xyz/dataset.”
For papers: pair the BibTeX entry with the access date and methodology URL agentcrush.xyz/methodology. The per-dataset-page canonical URL is stable.
Contact
Research collaboration, data questions, or custom exports: contact@agentcrush.xyz
Or on X: @agentcrush_xyz