Secrets are never stored
When a key or password turns up, only the rule name and an 8-character fingerprint are kept. The secret never reaches a log, a database or a report.
R-Secure runs on machines you own. This page sets out exactly what it records, where it keeps it, and the one small thing it sends out.
Pick a step, or let it play. This is what happens when an AI tool opens a file that holds a live key.
Claude Code reads ~/payments-api/.env on the developer's laptop. Its hook passes the event to the R-Secure agent on the same laptop.
Everything in the first column lives only on your R-Secure server.
There are three. Only one of them leaves your network, and it carries counts.
One event per action: tool, machine, project, path or command, rules matched, fingerprint. No secret values, no file contents.
InsideSign-in and dashboard pages, served from your own server.
InsideUsage counts only: number of agents, number of events, server version. Never file names, commands, prompts or code.
Outside// the whole license check: all RAKSYON receives { "agents": 12, "events": 450, "server_version": "1.4.2" }
It doesn't sit between your AI tools and their model providers. Claude Code and Cursor still talk to their own services, as they do today.
To keep a critical file out of a model's context, the hooks stop the read before it happens. For everything else, R-Secure records the access so your team knows exactly what each AI tool saw.
R-Secure sees the same sensitive files your AI tools do, so it's built to keep almost nothing.
When a key or password turns up, only the rule name and an 8-character fingerprint are kept. The secret never reaches a log, a database or a report.
File contents are checked on the developer's machine and thrown away immediately.
A key typed into a command is replaced with a redacted marker before anything is saved.
All activity data lives on your server. The license check carries counts, never content.
The hooks write one line and exit. If your R-Secure server is down, AI tools keep working normally.
Tampering with R-Secure's files or hook settings is itself flagged. On Linux, the kernel watcher runs as a protected system component the AI tools can't edit.
Older events are deleted from your server automatically. Saved daily reports aren't affected.
90 days by defaultRecord what developers asked, or turn it off for the whole organization.
On or offAdmin accounts with roles decide who can see activity in the dashboard.
Per personEvery developer machine has its own access token. Revoke one and that machine stops sending events.
Per machineAgents run on Linux (Ubuntu), macOS and Windows. The kernel watcher runs on Linux only.
Your machinesWe want to hear about security issues in R-Secure, and a person on the RAKSYON team reads every report.
security@raksyon.techContact: mailto:security@raksyon.tech Expires: 2027-05-04T00:00:00.000Z Preferred-Languages: en Canonical: https://raksyon.ai/.well-known/security.txt Policy: https://raksyon.ai/privacy.html#disclosureOpen file
Include the details. Steps to reproduce, the R-Secure version, and the platform: Linux, macOS or Windows.
Keep it private for now. Give us time to fix the issue before you share it publicly.
We'll work with you. We'll confirm the issue, keep you updated, and agree a disclosure date with you.
Don't send secrets. If a report needs a real key or customer data to show the problem, describe it instead.
A walkthrough with the RAKSYON team on your own infrastructure.