Lens
Inline policy enforcement for AI agents—monitoring every tool call, memory operation, and agent action in real-time with privacy-preserving telemetry and cryptographic audit trails.
"Transparency of process, not of content."
Govern AI Without Compromise
Lens solves the critical challenge facing enterprises: How do you govern AI systems without compromising privacy, performance, or innovation?
Policy Enforcement
Define and enforce ethics policies with a tiered response model: ALLOW → WARN → THROTTLE → DENY → QUARANTINE → TERMINATE → HITL. Semantic versioning, ethical citations, and sequence detection for multi-step behavior patterns.
Privacy-Preserving
Only anonymized metadata leaves your infrastructure—raw content never exposed. Governance without compromising data privacy.
Sub-Millisecond Latency
Fast-path evaluation adds less than 2ms p95 to agent operations. Production-grade performance without compromise.
Cryptographic Audit Trail
Tamper-evident Merkle chain records every decision. KMS-backed signing with AWS, GCP, Azure, or HashiCorp Vault.
Human-in-the-Loop
Escalate high-risk decisions to human reviewers in under 2 seconds. Configurable thresholds, approval workflows, and a dedicated HITL console to approve, veto, or escalate with full audit trail.
Framework Integrations
Native integrations for LangChain, LangGraph, LiteLLM, OpenAI, and Anthropic. Decorator-based instrumentation for custom tools, async support, and fail-open mode for production resilience.
Your Data Stays With You
The Lens SDK extracts features like size, destination, and type—raw content never leaves your environment. Only anonymized metadata reaches the control plane.
Your Infrastructure
- AI Agent + Lens SDK runs locally
- Extracts features (size, destination, type)
- Raw content NEVER leaves your environment
Lens Control Plane
- Policy evaluation on metadata only
- Audit logging with no raw data access
- HITL workflows for escalation
Policy Engine Features
Allowlisting permits fine-grained access to tools, parameters, domains, and HTTP methods of your choosing
Prevent access to tools, parameters, access levels, topics, content based on defined criteria
High-risk or sensitive actions trigger HITL workflows for escalation
Rule engine content comes mapped to multiple frameworks; laws, regulations, security, safety, ethics, and more
Rule chaining enables advanced logic for detecting and responding to potentially harmful or unsafe actions, operating within both deterministic and non-deterministic frameworks
Leverage advanced models to flag unsafe or harmful content in requests and responses in near-real-time
Production-Grade Speed
| Metric | Target | Actual |
|---|---|---|
| SDK added latency | ≤5ms p95 | <2ms |
| Sidecar policy eval | ≤2ms p95 | <1ms |
| Control plane eval | ≤30ms p95 | <15ms |
| Incident to HITL | ≤5s | <2s |
| Audit write | ≤10ms | <5ms |
Regulatory Alignment
Lens helps organizations meet requirements across major AI regulations with built-in policy citations and audit capabilities.
EU AI Act
- •Human oversight (Art. 14)
- •Transparency (Art. 13)
- •Risk management (Art. 9)
GDPR
- •Data minimization (Art. 5)
- •Accountability (Art. 5)
SOC 2
- •Monitoring controls
- •Audit logging
NIST AI RMF
- •GOVERN, MAP, MEASURE, MANAGE lifecycle
Canadian AIDA Act
- •Impact assessment filings
- •Explainability + risk mitigation duties
Cryptographic Audit Features
Tamper-evident record of every AI decision
AWS KMS, GCP KMS, Azure Key Vault, HashiCorp Vault
Every verdict linked to policy version and rules
Full transcripts and tool call logs for investigations
Cryptographic verification of audit chain integrity
Export audit data for regulatory review
Flexible Deployment Options
Cloud-Hosted (Multi-Tenant SaaS)
- • Fully managed control plane
- • Multi-tenant isolation
- • One-click deployment
- • 99.95% SLA
Secure Cloud-Hosted (Isolated SaaS)
- • Deploy in a private VPC
- • Single-tenant deployment
- • Full data isolation
- • Custom compliance requirements
Hybrid
- • SDK and sidecar on-premises
- • Control plane cloud-hosted
- • Best of both worlds
- • Flexible data residency
Cluster-Level Safety
As organizations deploy fleets of AI agents, governance must scale from individual agents to coordinated clusters. Lens provides comprehensive cluster-level safety.
Agents operate freely within policy boundaries but cannot expand their own authority. Authority hierarchy: international standards → org policies → policy engine → agent operations.
State-machine detection of multi-step behavior patterns across agents. Identify coordinated actions that individually appear safe but together pose risk.
Real-time max_risk, avg_risk, and min_risk across all agents in a cluster with force-directed interaction graph visualization.
ALLOW → WARN → THROTTLE → DENY → QUARANTINE → TERMINATE → HITL. Graduated enforcement with auto-escalation rules.
Behavioral baselines with automatic drift alerting. Detect persona drift, goal drift, and performance degradation across the cluster.
Fail-mode configuration (open/closed), container isolation with resource limits, dead letter queues, and automatic recovery with graceful degradation.
Formal Safety Specification
Lens includes a formal safety model specification covering the foundational guarantees that underpin every governance decision.
Trust Boundaries
Where data, control, and authority change hands
Authority Model
Who can do what, with layered human override
Isolation Layers
Tenant, agent, cluster, memory, and failure domain isolation
Escalation Paths
Tiered response from ALLOW to HITL with auto-escalation
Failure Containment
Blast radius minimization and deterministic recovery
Swipe to see all steps
Our Design Principles
Raw content never leaves your infrastructure
Every decision is auditable and explainable
Policies cite international standards and regulations
Drop-in integrations, not rewrites
Agents operate within defined boundaries; they cannot expand their own authority
Multiple independent safety layers, each providing independent guarantees
Complete with Insight
Lens and Insight together form Veilfire's comprehensive AI governance platform. While Lens provides real-time enforcement, Insight delivers comprehensive and continuous evaluation across five pillars before and during deployment.