Client privacy for AI coding work

Detect sensitive data. Help protect your clients.

Use AI coding tools with greater care for the information clients trust you with. PonoLens detects configured sensitive data locally, offers experimental redaction in supported tools, and gives solo developers, agencies, and small businesses a clear record to review and report.

For macOS 13 or newer · Works on Apple silicon and Intel · Saves activity on your Mac

Experimental Redact stops a supported prompt and provides a sanitized copy for you to paste and submit. Coverage depends on the tool and configured hooks. Check tool support.

Prompt HistorySaved on this Mac
Needs reviewCodex · just now
“Explain this customer error using their account details.”
What PonoLens found

An email address and customer account number.

Where it went

OpenAI, through Codex.

What happened

Observed after sending through a Codex session record. This observation could not stop it; local prompt hooks can provide experimental Block and Redact.

Why this helps: You can see the mistake, remove private details next time, and keep a redacted local record.

Who is it for?

For people trusted with other people’s data.

From your first freelance client to a growing team, make privacy checks part of your AI coding workflow.

Solo developers

Build client confidence.

Catch configured client details and secrets in supported AI activity. Review findings and export redacted reports to help explain your work when a client asks.

Agencies & consultants

Make client reporting easier.

Turn supported AI activity into clear, redacted reports for clients. Show what was observed, what needs review, and which actions a supported adapter actually blocked.

Small businesses

Handle sensitive work with more care.

Working in healthcare, legal services, finance, or another client-focused business? Check supported AI coding activity for configured personal, health, legal, and financial information to help protect your clients’ private data.

PonoLens supports privacy and compliance workflows but does not itself establish compliance. Detection can miss information; protection depends on supported integrations and your configured policy.

Detect. Redact. Report.

Practical privacy checks. Clear client reports.

Find configured sensitive information, take action where supported, and keep a readable record across connected AI coding tools. Detection runs on your Mac.

01

Detect sensitive information locally

Check supported prompts and actions for secrets, contact details, health, legal, and financial information. Add local exact values or custom patterns for the client information you need to watch.

02

Redact before sending, where supported

Experimental Redact stops the original prompt and gives you a sanitized copy to review, paste, and submit manually. Experimental Block stops supported actions. Observation-only integrations flag findings for review after or during an action.

03

Create redacted reports for clients

Keep redacted records in local storage on your Mac. Export redacted evidence when a client asks what happened. If you enable cloud archiving, redacted records are uploaded to your account for 365 days so you can review your registered computers remotely. Encrypted originals stay on your Mac.

How it works

Connect, check, and take action.

PonoLens works with supported AI coding tools. Some tools let PonoLens stop a risky prompt before it is sent. Other tools only let PonoLens show you what happened afterward.

01

Connect an AI tool

Connect Codex, Claude Code, Cursor, or Windsurf. PonoLens tells you what it can see and what it can stop for each tool.

02

Check for private information

PonoLens checks prompts and supported actions on your Mac. It looks for items such as passwords, email addresses, health details, and payment information.

03

Act on findings and share a report

Use experimental Redact or Block where supported. Review activity cards to understand what was observed, sent, or blocked, then export a redacted report for your client.

PonoLens vs. a free hook script

Can’t I just write a hook script?

Yes. If you use one AI tool, only need an API-key check, and are happy reading raw logs, you should. Hooks are great. PonoLens is built on hooks and supported local observers.

When a simple hook script stops being enough
Need Free hook script PonoLens
Tools & maintenance You connect each tool and maintain its script. Supported adapters maintained in one app.
Broader risks Only the checks you write or add. Configured checks for API keys, client names, emails, health and payment details.
What gets saved You choose what to log and redact. Sanitized audit records; saved prompts redacted by default.
A client asks what happened You interpret the output and prepare an answer. Redacted, plain-language records and exportable reports.

A custom script can add these features too. The difference is who builds and maintains the workflow.

PonoLens is honest about what it knows.

Observed

It recorded an action during or after it happened.

Sent

The integration indicates data left the device.

Blocked

A supported adapter actually enforced a block.

Redact is the default for new policies, with Report where unsupported. Existing saved choices are preserved. Redact and Block are experimental and need supported before-action hooks. Codex desktop and CLI support trusted synchronous local prompt hooks; cloud and session-observation paths are observation-only.

Detection isn’t perfect. Here’s exactly what’s stored, where, and what’s uploaded →

Read the short comparison → · Explore the demo

Simple, private pricing

Loading price… Solo · Loading price… per Small Business user.

Try every activity detail free for 30 days. After your trial, choose PonoLens for the current annual per-user price shown on Pricing.

Try PonoLens for Mac

Supported AI coding tools

What works with PonoLens?

The collector inspects supported hooks without an open dashboard. Unexpected collector outages, timeouts or invalid inspection stop supported synchronous submissions. An explicit Pause inspection, Stop Collector or Quit Collector allows continuation with a warning. Bypassed or unsupported hooks cannot guarantee inspection or recording. Use the menu-bar Start Collector or Restart Collector controls to recover; see recovery steps and harness limits.

Scanning stays local. Enable Scan files and its Binary files child for bounded document and image inspection, including supported PDF, Office and Apple document contents and local OCR. Some formats and file regions remain only partially inspected; read the file-scanning limits. The private v0.2.57 test build is separate from public downloads.

A check means PonoLens can see or do it when the tool is connected.

AI tool See prompts See tool activity Stop a risky prompt
Codex (Mac desktop / CLI) ✓ Yes ✓ Yes △ Testing
Claude Code ✓ Yes ✓ Yes △ Testing
Cursor ✓ Yes ✓ Yes △ Testing
Windsurf ✓ Yes ✓ Yes △ Testing
Devin Desktop ✓ Yes ✓ Yes △ Testing
Grok Build ✓ Yes ✓ Yes △ Testing
Muse Code ✓ Yes ✓ Yes △ Testing
Google Antigravity × No ✓ Yes × No
OpenCode ✓ Yes ✓ Yes △ Testing
Oh My Pi (OMP) ✓ Yes ✓ Yes △ Testing
OpenClaw ✓ Yes ✓ Yes △ Testing
Hermes Agent ✓ Yes ✓ Yes × No
macOS Harness × No ✓ Yes × No
Claude Code Desktop local Code
Private v0.2.57 · native window check pending
Testing · local Code runtimeTesting · supported hooksTesting · stop-and-copy
Claude Desktop Chat / Cowork × No × No × No

Testing means the feature is experimental. It works only when that AI tool gives PonoLens a supported before-send hook. Codex Block and Redact require trusted synchronous local prompt hooks. Redact stops the original and provides a protected copy to paste and submit manually. Auto redact can copy without a review modal. Cloud chats, untrusted or background hooks, and session observation cannot stop prompts. Coverage requires a compatible updated app and refreshed hooks; public downloads may contain an earlier release. Grok covers typed CLI prompts, Muse synchronous trusted CLI hooks, Oh My Pi interactive terminal input, and OpenClaw normal local-agent gates. RPC/print/ACP prompt protection in Oh My Pi and other untested runners are excluded. Grok and Muse native hook failures can allow continuation. Antigravity desktop, OpenCode CLI V1 and Hermes CLI provide supported tool blocking. Refreshed OpenCode CLI V1 chat.message hooks also provide experimental prompt stop/replacement and logged Skip; Desktop and V2 prompt enforcement remain unverified. Oh My Pi interactive input and OpenCode CLI V1 can apply Use redacted version after a logged choice; other supported hooks keep the original stopped for copying and manual submission.

If you sent something private

Stop, check it, and get help.

PonoLens can help you review what happened. It cannot pull back information that was already sent.

01

Stop sending more

Pause the AI task. Open the PonoLens activity card and check what may have been sent and where it went.

02

Replace secret keys

If you sent a password, API key, or access token, turn it off and make a new one right away.

03

Tell the right person

If it was customer, health, payment, work, or other private information, tell the person in charge and follow your incident steps.

You can also ask the AI provider or your account owner to delete the chat. Deleting it may help limit later access, but it does not undo the original send.

Testimonials

Testimonials

Questions and answers

Ask about PonoLens.

Get a simple answer from the PonoLens help pages.

Your question stays on this page. It is not sent to an AI company.

About us

Built to make AI coding activity understandable.

PonoLens started as a weekend side project and has grown into a focused Mac app for people using AI. The idea remains practical: make supported coding activity understandable without asking people to read technical logs.

Holmgren Studio, LLC continued building PonoLens into a full clear activity viewer for a future where AI is part of everyday work. We believe it should explain activity clearly and keep the audit history it protects on your Mac.