Before a tool touches a codebase, someone eventually asks the only question that matters: where does our code go now? For Power Claude the answer is short — nowhere new. It runs inside your editor, reads your local session files, and never receives your prompts or your code. Here is exactly what leaves your machine, and what doesn't.
The question a security review actually asks
Adding any extension to a developer's editor reopens a question that procurement, security, and your own conscience all care about: does this change where our source code and our prompts end up?
It is a fair question, and most "privacy" copy answers it badly — usually with a sweeping claim that nothing ever leaves your machine. That claim is easy to write and easy to disprove, because the moment you use any hosted model your prompts go to that model's provider. Overstating privacy is worse than saying nothing: it invites the reader to trust a boundary that isn't there.
So here is the precise version for Power Claude, framed around what we — Neural-LLM — can and cannot see.
What Power Claude is, and what it is not
Power Claude is a companion that runs alongside Claude Code inside VS Code, Cursor, or any compatible editor. It manages your own Claude accounts: it rotates between them when one hits a limit, watches for crashed or stalled sessions, and reads the local session files Claude Code already writes to render a dashboard.
Two things follow from that design:
- There is no Neural-LLM server in the request path. Your prompts travel the same route they did before you installed anything — from Claude Code on your machine to your own Claude account. Power Claude decides which of your accounts a request uses; it does not stand between you and the model as a relay, and it does not copy the traffic.
- The dashboard reads, it does not upload. Account utilization, rotation history, token burn, and session state are all derived from local files in
~/.claude/and~/.power-claude/. They are rendered in the editor. None of it is sent to us.
What Power Claude is not: it is not a proxy that ingests your conversations, not an analytics product that phones home with your usage, and not a model provider. It never sees, stores, or transmits your prompts, your code, or Claude's replies. We did not build the endpoint that would receive them, because the product does not need one.
The one thing that leaves your machine
Being concrete matters more than reassurance, so here is the complete list of what Power Claude sends to Neural-LLM: a license check. It carries a hashed license key and a machine identifier so a paid seat can be validated, plus a timestamp. It is cached, so it happens roughly once a day, not once a request.
That request contains nothing about your work — no prompt text, no file contents, no account credentials, no record of what Claude said. Turn off your network entirely and Power Claude's local features keep working from cache; the only thing that degrades is the periodic license validation.
This is the whole reason the privacy story is simple. A tool that never collects something cannot leak it, cannot be subpoenaed for it, and cannot have it exposed in a breach. The smallest attack surface is the one that was never built.
What this does and does not mean for compliance
It is tempting to translate "we never see your code" into a compliance badge. It isn't one, and pretending otherwise would be the same overstatement this article opened by criticizing.
Here is the honest mapping. Power Claude does not change your data path, so it does not change your obligations. Whatever rules already govern your use of Claude Code on a given codebase — a client agreement, an internal data-handling policy, a regulatory regime — apply exactly as they did before, because your prompts still go to the same place under the same account. What your prompts and code are subject to once they reach Claude is governed by your own agreement with Anthropic. That part sits between you and Anthropic and is outside Power Claude's scope.
What Power Claude can say cleanly is narrower and more useful: it adds nothing new for a reviewer to assess. The answer to "where does our code go with this installed?" is identical to the answer without it. For a security questionnaire, "introduces no new data flows and no new vendor in the request path" is a real, checkable statement — and it is the one we make.
How to verify it yourself
You do not have to take the architecture on faith. A few ways to confirm it:
- Watch the network. Run Power Claude with a local proxy or packet inspector and you will see the periodic license check and your own Claude traffic — and no third stream carrying your prompts to a Neural-LLM host.
- Read the local files. Everything the dashboard shows comes from
~/.claude/and~/.power-claude/. Delete that directory and the history is gone; there is no server-side copy to restore from. - Pull the plug. Disconnect from the internet and the rotation, watchdog, and dashboard keep running on local state. A product that needed to upload your data would not survive going offline.
The point of a privacy claim is that it should be falsifiable, and this one is. Verify it, don't trust it.
Frequently asked questions
Does Power Claude send my prompts or code to Neural-LLM?
No. Power Claude runs inside your editor and has no server in the request path. It never sees, stores, or transmits your prompts, your code, or your conversation. The only thing it sends us is an anonymous license check.What exactly is in the license check?
A hashed license key, a machine identifier, and a timestamp — enough to validate a paid seat. It contains no prompt text, no file contents, and no account credentials, and it is cached so it runs about once a day.Does using Power Claude make my project compliant with a standard like PCI or SOC 2?
No. Power Claude is a local companion tool, not a compliance certification, and it does not claim to be one. It does not change your data path, so your existing obligations are unchanged; what happens once your prompts reach Claude is governed by your own agreement with Anthropic.Can I confirm none of my code is leaving?
Yes. Inspect the network traffic and you will see only your own Claude account requests plus the periodic license check. Power Claude's data lives in local directories you can read and delete, and its core features keep working with the network disconnected.Start with both ways in
If "no new data flows" is the property you need, Power Claude is built to be checkable rather than to be taken on trust — see how the local-only architecture works on the product page. When you are ready, start the 14-day Premium Trial at neural-llm.com/pricing — card on file, $0 today, and it locks founder pricing for two years. Prefer no card up front? Install Power Claude free and the no-card 7-day trial activates the moment you open the extension.