// security & trust
Your code never leaves your machine.
That is the architecture, not a promise.
Ghost Architect™ is a command-line tool that runs entirely on your laptop, your server, or your CI runner. No Ghost Architect™ server ever sees your source. There is no central server holding your scans. The product is designed so that even if Ghost Architect™ were compromised tomorrow, your codebase would not be affected. That holds for both products: with Ghost Open™ the model is Anthropic’s, reached with your own API key, and never ours; with Ghost Architect™ Local the model runs on your hardware too, and nothing leaves your network.
Ghost Open™ · free for one person
The scanner runs on your machine. The only outbound traffic is the request to Anthropic, under your own API key. We operate no server that sees your source.
Ghost Architect™ Local · your hardware NEW
The model runs on your machines, inside your network. One licence activation, then nothing calls home. An egress inventory generated from the code, a hash-chained audit log, a compliance ledger your auditor can read.
- ✓
Local execution
The scanner runs on your machine. No code is uploaded to Ghost Architect™.
- ✓
BYOK Anthropic API
You bring your own Anthropic API key. We never see it.
- ✓
No central storage
We operate no servers that hold customer source code.
- ✓
Reports stay yours
Reports are written to your own machine only.
- ✓
Telemetry is opt-out
Set
GHOST_NO_PING=1to disable all outbound calls. - ✓
Ghost Architect™ Local
The model runs on your hardware too. Zero egress, refused at the code level, with the inventory that proves it. Details ↓
- ✓
// data flow
Where your code actually goes during a scan
When you run Ghost Architect™ against a codebase, here is the complete path the data takes. There are no hidden steps.
- Step 1
Ghost Architect™ reads your local files. Nothing is uploaded yet. File reading happens entirely on your machine.
- Step 2
Relevant code chunks are sent directly to Anthropic's API using your own API key. The traffic goes from your machine to Anthropic. Ghost Architect™ is not a relay, proxy, or middle hop. Anthropic's enterprise data retention policies apply (zero data retention by default on paid plans). See Anthropic's privacy policy.
- Step 3
Anthropic's API returns analysis. The response comes back to your machine.
- Step 4
Ghost Open™ writes the findings report locally as PDF, Markdown, or text, on your machine only. Nothing is published anywhere.
- Step 5
A small heartbeat ping containing the CLI version, an anonymized install ID, and the event type (for example
cli-firstrun-docker) may be sent to our telemetry endpoint. This ping contains no source code, no file paths, no findings, no API keys, and no personally identifiable information. You can disable it entirely by setting the environment variableGHOST_NO_PING=1.
// architecture
What we operate, and what we do not
- We operate
A static marketing website (this one), a Cloudflare Worker that receives anonymous heartbeat pings for install analytics, an Airtable base that stores those anonymous pings, and the Ghost Open package on the public npm registry.
- We do not operate
Any server that stores, processes, or transmits customer source code. Any database of customer findings reports. Any AI inference infrastructure that sees your code. Any analytics service that profiles your team or users.
- Authentication
Your Anthropic API key is stored locally on your machine in a configuration file you control. It is never transmitted to Ghost Architect™. You can rotate or revoke it at any time without involving us.
// compliance posture
On SOC 2 and ISO 27001
Ghost Architect™ is operated by a solo founder running a deliberately minimal infrastructure footprint. We do not currently hold SOC 2 Type II or ISO 27001 certifications. We believe in being direct about that rather than displaying badges we have not earned.
The reason we are comfortable engaging with security-conscious teams without those certifications is that the product is architected so that the data those frameworks are designed to protect never enters our infrastructure in the first place. Your source code, your API keys, and your findings reports all stay on systems you control. There is no Ghost Architect™ data plane for an auditor to certify.
When Ghost Architect™'s enterprise revenue justifies the audit cost, and when an enterprise customer requires it as a condition of contract, we will pursue SOC 2 readiness through a recognized provider. Until then, we direct your security team to this page, to Anthropic's enterprise compliance documentation, and to the source code itself.
// ghost architect local
Zero egress, at the code level, with the inventory that proves it
Everything above describes Ghost Open™, where the model is Anthropic's, reached through your own API key. Ghost Architect™ Local is a different deal: the model runs on your hardware too. Your source never leaves your network, because there is nothing on the other side to send it to.
- One call home, once. Licence activation sends the licence key and an install id to ghostarchitect.dev and receives a signed licence. Nothing from the codebase is in the request. The licence is verified on the machine with Ed25519; there is no phone-home to keep running.
- An egress inventory, not a policy page. The product prints every destination it is capable of sending to (email, webhooks, team sync, PR comments, a git server), what each one would carry, and the switch that turns it off. The inventory is generated from the code that gates the traffic, so it cannot drift from what the product does.
- Zero-egress mode refuses, it does not warn. Turn it on and every outbound integration is refused at the code level, including a spoke enrolling against a hub outside your LAN. The refusal is written to the audit log.
- A compliance ledger your auditor can read. Every scan ending, completed, failed, cancelled, killed or lost, is a row with the scan it belongs to, the person, the project label and the directory name (never the full path). Exports are CSV or PDF, signed, with the signing key rotated from Admin Console.
- A hash-chained audit log. The audit log is chained line by line and the Audit page verifies the chain inside every file it opens, and says which files it opened, so "the chain verifies" and "no file is missing" are two claims kept apart on purpose.
- Reports that name their coverage. A report says how many files it read of how many exist and claims nothing about the rest. A scan that never finished counting says nobody counted, never zero.
- Redaction before the model sees it. Credentials, keys, connection strings and tokens are redacted from the text the model receives, and the redactor fails closed: a file it cannot redact is not sent.
- Verify it yourself. Run it behind a packet inspector with zero-egress on. After activation you will see LAN traffic between your own machines and nothing else. The product ships as source in the tarball you install, so the code that makes these claims is the code you can read.
// for security reviewers
Verifying these claims yourself
Everything described on this page is independently verifiable.
- Inspect the source. Ghost Open™ is installable from npm as
ghost-architect-open, free for one person and source-available under the Business Source License 1.1. Read the package contents before you run it. - Watch the network. Run Ghost Architect™ behind a packet inspector. You will see two endpoints: the Anthropic API and our telemetry endpoint. Nothing else.
- Disable telemetry. Set
GHOST_NO_PING=1in your environment. All outbound calls from Ghost Architect™ itself stop. Only Anthropic calls remain, and those use your key. - Run it air-gapped. Ghost Architect™ will run on an isolated machine with only Anthropic API access. No connection to Ghost Architect™ infrastructure is required for scanning to function.
Security team has questions?
We answer security review questionnaires directly. No vendor portal, no compliance theater.
Last updated: October 5, 2026