See what your agents saw.

One command scans your AI session logs locally, surfacing security risks, PII exposure and unauthorized access before they become incidents. Open source, runs locally, no signup.

View on GitHub (opens in a new tab)Get started free
install with Homebrew
$ brew install hoophq/tap/hooprs
# prebuilt, no compile step; npm also covers Windows
# then scan every local session:
$ hooprs

What a scan looks for.

Secrets, financial data, national IDs, health and contact details, each with a severity. The highest severity in a session sets its tier.

Sessions by risk

Every session lands in one tier. Pick a tier to see the entity types that put it there.

48 entity types · 9 families

  • SecretAPI_KEY, high severityAWS_ACCESS_KEY, high severityPRIVATE_KEY, high severityPASSWORD, high severity
  • FinancialCREDIT_CARD, high severityUS_BANK_NUMBER, high severityIBAN_CODE, high severityCRYPTO, high severityABA_ROUTING, medium severityAU_ABN, low severityAU_ACN, low severityIN_GSTIN, low severityIT_VAT_CODE, low severitySG_UEN, low severity
  • Government IDUS_SSN, high severityUS_ITIN, high severityUK_NINO, high severityAU_TFN, high severityIN_AADHAAR, high severityIN_PAN, high severityIT_FISCAL_CODE, high severityES_NIF, high severityES_NIE, high severitySG_FIN, high severityPL_PESEL, high severityKR_RRN, high severityFI_PERSONAL_IDENTITY_CODE, high severityTH_TNIN, high severityUS_PASSPORT, medium severityUS_DRIVER_LICENSE, medium severityIN_PASSPORT, medium severityIN_VOTER, medium severityIT_PASSPORT, medium severityIT_IDENTITY_CARD, medium severityIT_DRIVER_LICENSE, medium severity
  • HealthUK_NHS, high severityAU_MEDICARE, high severityMEDICAL_LICENSE, medium severityMEDICAL_RECORD, medium severity
  • ContactEMAIL_ADDRESS, medium severityPHONE_NUMBER, medium severity
  • NetworkIP_ADDRESS, medium severityURL, low severity
  • IdentityPERSON, low severityLOCATION, low severityNRP, low severity
  • ContextDATE_TIME, low severity
  • OtherIN_VEHICLE_REGISTRATION, low severity

From the tool’s own catalog. PERSON, LOCATION and NRP need the opt-in model (-ner); every other type is pattern and checksum matched.

One command, three answers.

It finds the sessions of Claude Code, Cursor and OpenCode on disk and runs detection in-process.

  1. Install and run

    One command discovers every session and prints a risk summary.

    hooprs

  2. Get your score

    A security score from 0 to 100, with every session tiered critical, minor or low.

  3. See your exposure

    Entity types, counts and the most exposed sessions, in one self-contained HTML report.

    risk-report.html

Private by design.

A scanner that reads your secrets should never send them anywhere. This one only writes to your disk.

narrow a scan
# only the last 30 days, plus a machine-readable report
$ hooprs -days 30 -json risk-report.json
# only Cursor sessions for one project
$ hooprs -tools cursor -project my-app
# print values masked to their last 4 characters
$ hooprs -mask-values

  • Fully offline

    Detection runs in-process. The default run makes no network call, so no session content ever leaves your machine.

  • Value-free reports

    The HTML and JSON reports carry entity types, counts and severities, never the matched values. Sharing one never re-leaks the leak.

  • Open source

    MIT licensed. Inspect every line of code before you run it.

More from hoop Labs.

Each tool solves one problem, runs locally and needs no signup to try.

See every labs tool

Every session, not a snapshot.

Risk Summary shows what your agents touched after the fact. hoop.dev sits in every session, recording it, controlling access and redacting PII before it ever leaks. No exports, no separate tool.

Get started freeSee our GitHub (opens in a new tab)

Ask an assistant what hoop.dev does: Claude, ChatGPT, Grok, Gemini, Perplexity, Copilot.

install and scan
$ brew install hoophq/tap/hooprs
$ hooprs