Product
From hidden signals to deployable lenses.
NeuronLens turns internal model signals into three deployable products: agent oversight, safety monitoring, and targeted model repair.
- Agent Lens controls agent tool calls from inside the model.
- Safety Lens detects grounding risk and policy violations before delivery.
- Model Repair Studio fixes recurring failures with targeted internal repair.
Three products. One research engine.
Agent Lens
Turn internal signals into live controls for agent tool calls and behavior.
Safety Lens
Detect grounding risk and policy violations from inside the model.
Model Repair Studio
Fix recurring model failures with targeted internal repair.
01 — Agent Lens
Agent Lens: internal oversight for tool-using AI
Connect in 2 lines · works with LangChain, CrewAI, Claude Agents, AutoGen, OpenAI Agents, and more
Finds model-internal signals that drive agent behavior
Catches risky tool calls before execution
Detects tool-call hijacking, goal drift, and privilege escalation
02 — Safety Lens
Safety Lens: internal signals for grounding and policy risk
Detects grounding and policy risk while responses form
Catches prompt injection, RAG poisoning, unsafe output, and red-team patterns
Supports block, review, reroute, or steer, with evidence before delivery
03 — Model Repair Studio
Repair failures from inside.
When the same failure keeps appearing, Model Repair Studio helps diagnose the internal features behind it and guide targeted repair — without blunt retraining.
Failure Slice
Group recurring errors into a focused repair set.
Feature Diagnosis
Identify internal features associated with the failure.
Targeted Repair
Guide adaptation toward the failure-driving signals.
Before / After Evidence
Measure whether the repair fixed failures while preserving what already worked.
Fix the failure mode, not the whole model.
One loop. Three products.
The same research engine powers all three products: activation analysis, sparse features, probes, and causal validation.
Control
Agent Lens
Internal oversight for tool-using AI agents. Catches risky tool calls before execution.
Protect
Safety Lens
Detects grounding risk, prompt injection, and policy violations from inside the model.
Repair
Model Repair Studio
Diagnoses and fixes recurring failures with targeted internal repair.
Start with one workflow. Build reusable lenses.
NeuronLens is looking for design partners deploying high-stakes AI systems where surface monitoring is not enough.