Loading...
Loading...
Loading...
Every badge on this site links here. Benchmarks we run, scans we run, identity we verify, and a trust score whose formula is public — evidence over assertion, and paid placement never touches any of it.
Verification binds to a specific agent version. A new version drops the listing back down the ladder until it is re-certified.
Drafts and pending submissions. Not visible in the directory.
Every listing carries an explicit L0–L4 classification, decided by a reviewer from observed benchmark behavior — not from marketing copy. Listings may not claim a higher level than their classification.
Responds to prompts. No tool execution.
Executes tools, but every action is human-triggered or human-confirmed.
Executes a predefined multi-step sequence with AI steps inside; cannot re-plan when steps fail. Most products marketed as “agents” today are L2.
Plans, selects tools, retries and re-plans within a scoped objective; maintains state; escalates at policy boundaries.
L3 plus long-horizon operation, cross-system state, self-monitoring, and full decision audit logs.
Fees fund sandboxed benchmark runs, the scan pipeline, and human review. This table is live configuration — changes appear here immediately.
The full versioned methodology covers benchmark scoring, all eight scan types, the verification lifecycle, and the exact trust score formula.