It deletes the wrong thing
One confused step and important files or records are gone — with no undo button.
Today's AI assistants don't just chat — they can delete files, send your data, and spend your money. Siphrix watches what they try to do and quietly blocks anything risky before it happens. No code needed.
Think of it as a seatbelt for AI — you hope you never need it, but you're very glad it's there.
The same assistant that books your travel can also make a mistake you can't take back. Here's what Siphrix is built to stop.
One confused step and important files or records are gone — with no undo button.
Passwords, customer data or personal documents quietly sent somewhere they shouldn't go.
A sneaky webpage or email tells the AI to misbehave — and it follows the instructions.
An over-eager assistant makes a purchase or transfer you never approved.
You can't watch every move an AI makes. Siphrix can.
Siphrix sits quietly between your AI and the real world. It checks every action before it can happen.
Your assistant decides to send a file, run a command, or make a change.
It compares the action against your safety settings — in a fraction of a second.
Harmless actions just happen. Dangerous ones are blocked — and you're told why.
Choose how protective you want to be, then pick an action. This runs right here in your browser — nothing actually happens to anything.
No settings to wrestle with, no code. For each AI app you use, pick one of three levels in plain English. Change it anytime.
Pick a level per app · change it whenever you like
Connect every browser, computer or AI agent to your account. See where each one is, whether it's online, and set how protective to be — for each, from anywhere.
Change a setting here — that device picks it up in seconds. Your data stays on each device; only the rules travel.
Siphrix runs on your own computer. It doesn't phone home, and your information never leaves with it.
If something looks risky or unclear, Siphrix stops it rather than guessing. Better safe than sorry, always.
Every decision is written down in plain terms, so you can always see what happened and why.
Under the friendly surface, Siphrix is a full policy engine you can drop into any agent — with a Python SDK, a CLI, an MCP server and editor plugins. It's local-first and fail-closed by default.
evaluator.evaluate(action, ctx)pip install siphrix && siphrix demopython -m siphrix.mcpVS Code · Claude Code · CodexNot at all. You just pick a protection level for each app in plain language. The technical depth is there if you want it, but it's completely optional.
No. Each check takes a fraction of a second, so you won't notice it — you'll only notice when it saves you from a mistake.
Risky actions an AI might take on your behalf: deleting files, sending private data, spending money, or being tricked into harmful instructions.
No. Siphrix runs locally on your machine. Your data and the things your AI does stay with you.
The action simply doesn't run, and Siphrix tells you what it stopped and why. Nothing happens behind your back.
Anytime. Turn protection up or down per app whenever you like — it takes a couple of clicks.
Set it up in under a minute and let Siphrix watch your back.