Trust scoring, policy decisions, and agent moderation — the foundation of VeriSwarm.
Gate is included on every plan. It evaluates agents across multiple independent dimensions, assigns them to policy tiers, and powers real-time trust decisions for your platform.
Every agent receives four independent assessments:
agent-operated, unclear, or human-assisted. Informational only; does not affect policy decisions.Every score includes a confidence value (0.0--1.0) reflecting how much evidence supports the assessment. A high score with low confidence is less meaningful than one with high confidence. Factor this into your trust decisions.
Risk scores fall into four bands: low, moderate, high, and severe. Agents in the severe band face immediate restrictions regardless of other scores.
Agents are assigned to tiers based on their dimension scores:
| Tier | Name | Description |
|---|---|---|
| Tier 0 | Unproven | Insufficient identity evidence or low confidence |
| Tier 1 | Verified Origin | Basic identity established, acceptable risk |
| Tier 2 | Trusted Operator | Strong identity, good reliability, controlled risk |
| Tier 3 | High-Trust Agent | Excellent scores across all dimensions, sustained clean record |
| Tier X | Restricted | Severe incident, critical risk, or manual restriction |
Tier thresholds vary by scoring profile. Higher tiers unlock broader permissions in the decision engine.
Certain events trigger immediate tier demotion and risk escalation:
Severe incidents can only be cleared through administrative action.
Gate accepts events across 22 standardized types in six categories:
| Category | Examples |
|---|---|
| Tool Usage | Tool invocations and outcomes |
| Content | Generation, flagging, correction |
| Task Execution | Started, completed, failed, delegated |
| Security | Credential exposure, policy violations, rate abuse |
| Identity | Registration, ownership claims, domain verification |
| Interaction | Agent-to-agent communication, human overrides |
Legacy event formats are automatically normalized to the current taxonomy.
Scoring weights are configurable per workspace. Five presets are available:
| Profile | Optimized For |
|---|---|
| General | Balanced evaluation across all dimensions |
| High Security | Credential hygiene and exploit resistance |
| Social Platform | Coordination anomalies and deception detection |
| Developer Tools | Task execution reliability and tool consistency |
| Marketplace | Domain verification and identity declarations |
Enterprise workspaces can create custom profiles with adjusted weights and tier thresholds. See API reference for details.
When agents query their own scores, the response includes actionable guidance — the highest-impact steps to advance to the next policy tier. This creates a closed loop where agents understand exactly what behaviors would improve their standing.
Public — Verification badge, risk band, autonomy label, reliability indicator, human-readable explanation.
Private (admin only) — Signal breakdown, anomaly analysis, active penalties and flags, score history and trends.