Inspector — deep QC for AI agentsTry the live demo ↗

Inspector/deep QC/self-hostable

Your agent says it did the thing. Prove it.

Inspector audits another AI agent. It reads the agent’s real configuration, designs an exam from what the agent claims to do, holds six isolated conversations with it, diffs the data around every one, and reports which of its actions actually happened — and which it only said happened.

The demo needs nothing from you · ten seconds · free

Layers
5, from x-ray to forensics
Rubric
17 dimensions in four bands
Write verdicts
7, decided by evidence
Accounts
None. No login, no user database
01

The problem

A confident agent and a working agent read identically.

An agent that files your ticket says “I’ve created your ticket!”. An agent that files nothing says exactly the same sentence, in the same warm tone, just as fast. The customer believes it. The operator reading logs cannot tell. The request is simply gone, and nobody downstream ever learns it was dropped.

Asking the agent whether it worked is worthless — that is the one question a broken agent is guaranteed to answer wrongly. Inspector does not ask.

02

How it works

Five layers.

Each one narrows what the agent could be getting away with. The deterministic parts — the verdicts, the failure classifier, the credential scan, both rating caps — are plain functions with no model in the path.

LAYER 1

X-ray

read the target’s real config and tool inventory

LAYER 2

Exam design

generate scenarios, then repair them deterministically

LAYER 3

Live run

invoke the target, one isolated thread per scenario

LAYER 4

State diff

photograph the data before and after every scenario

LAYER 5

Forensics

read logs and traces, classify failures by owner

03

Why it holds up

Three things that do not depend on trust.

The canary

It proves the action happened.

A unique reference code is planted inside every write test, then requested back in a brand new conversation. An agent that really wrote something can find it. One that only said so cannot — and that is a phantom write, caught with no database access at all.

The caps

A generous grade cannot survive the evidence.

The rating caps are plain if statements applied after grading, never trusted to the grading model. A proven phantom write caps the agent at 2.0; a leaked credential caps it at 1.0. The model does not get a vote.

Null is an answer

It says what it could not see.

No logs, no config, no data store — the audit still runs, and every finding is stamped VERIFIED, CORROBORATED or CLAIMED according to what actually backed it. The report states its own blind spots and what access would close them.

04

How to use it

Four things to hand it.

Only the first two are required, and the second one is a text file.

01 · Point it at the agent.
A plain HTTP endpoint, an OpenAI Assistant ID, or a Lua agent ID. The endpoint only has to take a message and return a reply — Inspector sends a superset body and reads a superset of reply shapes, so most agents work with no adapter written.
02 · Give it the spec document.
What the agent is supposed to do: purpose, capabilities, tone. Paste it or upload a .md file. This is the half of the audit that finds gaps — a capability promised in the doc and missing from the agent is a finding no amount of conversation would surface.
03 · Optionally, hand it read-only telemetry.
Config, logs, traces or data access. Entirely optional; the audit runs without any of it. Supplying it is what turns the agent’s own word into proof, and the report is explicit about which findings were affected.
04 · Read the report.
A rating out of five, a headline risk, seven write verdicts, a 17-dimension rubric and a prioritised fix list with an owner against each one. Printable to PDF as real selectable text.
Never seen it before? The demo runs the whole thing against a bundled agent that lies about its own work — no endpoint, no key, no spec doc needed from you. Run the demo →
05

Run it

Audit your own agent.

Credentials arrive in the request, are passed to the adapter, and die with it. They are never written to the store, never included in the saved report, and never logged.

* This instance is demo-only. It has no AI provider key configured, so a real audit will not complete here — the form below is the real thing, but it needs a key to grade against. Self-host Inspector with your own ANTHROPIC_API_KEY to audit a live agent, or run the demo to see a full report right now.
01 · What are we auditing?
02 · The agent

Held in memory for this run only. Never written to disk, never included in the saved report, never logged — including on error paths.

03 · The spec document

What the agent is supposed to do — purpose, capabilities, tone. Inspector reconciles this against what the agent really has and really does; the gap between the two is where most findings come from. Paste it, or upload a .md / .txt file.

04 · Read-only telemetry — optional

The audit runs without any of this. Supplying it is what upgrades findings from the agent’s word to proof: config access turns “the doc promises refunds” into “no refund tool exists”, and data access turns a CORROBORATED phantom write into a VERIFIED one.

Try the demo instead

Roughly three to five minutes · keep the tab open

Inspector · self-hostable · no accountsTry the live demo ↗