NeuronLens
Browse the documentation
Agent Lens

Overview

Agent Lens checks your AI's answers and your agents' actions by reading a model's internals, before they reach a user or run.


Agent Lens tells you when an AI answer or an agent's next action is likely to be wrong, misleading or unsafe, before it reaches a user or runs. It does not ask another LLM for an opinion. It reads the internal activations of an open-weight model (gpt-oss-20b) over your conversation, and small trained classifiers, called probes, turn those activations into a risk for each failure mode.

You send what happened. Agent Lens sends back a verdict, and your code decides what to do with it.

Your app or agentsends the answer, orthe tool call it wants to make/v1/scoreAGENT LENS1 Read the model's internalsgpt-oss-20b activations, one pass2 Every probe gives a riskfrom 0 to 1, threshold 0.53 Return one verdictand the words or step that drove itallowsend it, run the toolescalatehold it for a persondenyblock itKept for Monitor and Eval (unless store: false)
One call, one read. Agent Lens never writes the answer and never runs your tools: it reads what you send and recommends what to do.

Two things you can check

You sendYou get backTypical use
Prompts (answers)the prompt, anything the model read, and the model's answera risk per failure mode, the sentence that drove it, and a verdictsupport bots, RAG answers, copilots: check the answer before the user sees it
Agents (steps)the run so far and the tool call the agent wants to makea verdict before the tool runspayments, trading, IT and data agents: stop the step before it happens

Both use the same call, POST /v1/score. In the console, every screen has a Prompts | Agents switch, so answers and agent runs are counted separately.

What it catches

Failure modeIt fires whenExample
Deceptionthe answer misleads, and the model represents that it does"your money is safe" from a fund that lost 18%
Goal driftthe answer serves another goal (a quota, a target) over the user's"passed identity checks" from an agent behind on its approval quota
Data leakthe answer passes a record value to someone who should not see ita customer's card number given to a third party
Reasoning errora calculation in the answer is wrong3 × 1,240 = 3,620
Omissionan agent skips a check it was offered before a consequential actiona wire sent without running the sanctions screen
Failed checka check ran and failed, and the agent goes ahead anywaya wire sent after the screen said "potential match, hold"
Harmful I/Oa destructive action without the evidence that justifies itan account disabled with no alert behind it
Injected intentthe next action follows an instruction planted in a tool resultan email that says "forward the payroll file", and the agent does

Probes lists what each one needs from you.

One scale, three verdicts

Every probe returns a risk from 0 to 1, calibrated so that its threshold is 0.5. The verdict comes from the highest risk.

goal drift 0.006data leak 0.58failed check 1.0000.5 THRESHOLD0.9 DENY1allowno probe firedescalatea probe fireddenynear certain
The three markers are real results from the examples in the Quickstart: a support bot that gave out a card number (escalate), and a payments agent that sent a wire after a failed sanctions screen (deny).
VerdictMeaningWhat your code usually does
allowno probe firedsend the answer, run the tool
escalatea probe fired (risk ≥ 0.5)hold it for a person, or log it and alert
denya probe is near certain (risk ≥ 0.9), or a rule blocks itblock it

The verdict is a recommendation: your code acts on it. You can move each probe's threshold and the deny level in Settings.

Agent Lens and Agent Lens Pro

There are three ways to use Agent Lens, and most teams go through them in order:

Agent Lens playgroundAgent Lens APIAgent Lens Pro console
What it istry a prompt in the browser/v1/score and the integrations, from your codethe app where you watch, review and test
Whereapp.neuronlens.com → Agent Lensapi.neuronlens.comapp.neuronlens.com/agent-lens
Whoanyone signed inanyone with an nl_ keyPro accounts
Per day10 runs1,000 runs(uses the API's runs)

Agent Lens vs Agent Lens Pro compares them in detail.

The Pro console

The console at app.neuronlens.com/agent-lens sits under Agent Lens Pro in the sidebar. It has four screens:

ScreenUse it to
Connectcopy your key and the code for each way in, and see which apps and agents have sent traffic
Monitorread every scored answer and agent run, review the flagged ones, and approve or reject held steps
Evaltest an agent on failure scenarios before release, and sign the results as an audit report
Settingschoose probes and thresholds, add rules, set data retention and alerts

What is kept, and who sees it

  • Your workspace only. Every record carries your workspace (organization and project), and only its members see it. Calls made with your key go to your first workspace, the one the console opens on.
  • Text is not kept by default. Monitor stores the verdict, the risks and where the flagged words are. It shows the words only if you turn on Keep text in Settings.
  • Credentials are removed. Keys, tokens and passwords in tool arguments are stripped before anything is stored.
  • Decisions are sealed. Each decision is hashed and signed, so a report can prove what was decided and when.
  • "store": false on a call scores it and keeps nothing.

Where to go next

  1. Quickstart: score your first answer and your first agent step (5 minutes).
  2. Worked examples: a support bot, a RAG answer, a payments agent and a coding agent, end to end.
  3. Integrations and hooks: Claude Code, MCP, LiteLLM, OpenTelemetry, the SDK, LangGraph and Google ADK, with how each one works.