Claude Code
In the terminal or VS Code, on the developer’s own Claude plan.
Every Claude Code prompt passes your AI policy before it leaves. Moring AI blocks secrets,
masks customer data, and gives your security team a record of every session.
Every Claude Code request is checked against your policy before it leaves. The Moring AI gateway runs in your own AWS account. Your data never passes through Moring AI. A redacted value is replaced with a label such as [REDACTED:aws_access_key_id], so Claude still understands the request.
In the terminal or VS Code, on the developer’s own Claude plan.
Checks the request against your AI policy. Secrets and customer data are blocked, redacted, or logged.
Claude sees only what your policy allows through.
Every request is checked before it leaves. Only approved models and Claude Code get through.
Every request is kept, cleaned: an audit trail you own, with no secret stored in it.
If a check can’t run, the request is refused. It is never sent unchecked.
About this much added to a typical request.
Requests handled with no errors in load tests.
About this long until a request shows in the Security Console.
Measured in load tests and the dev environment. Not customer production figures.
Your developer types one line. Behind it, Claude Code attaches the whole conversation: the files it read, command output, even .env files with passwords and keys.
Secrets and customer data leave the company with ordinary prompts, and nobody notices. You can’t see what was sent, you can’t stop it, and you keep no record to show an auditor or a customer.
Banning AI isn’t the answer. Trusting each developer isn’t enough either, because Claude Code reads files and runs commands by itself.
The Security Console shows your security team each Claude Code session, each person, and each policy decision.
Every Claude Code conversation, with what the policy did on each request. "Found in" shows whether a value came from what the developer typed or what Claude Code added by itself.
One row per person with sessions, requests, models, data sources, and cost. It flags who needs a look.
Start from the Moring AI default in monitor mode, then switch to enforce when you are ready.
Nov 11, 9:23:37 PM → 9:23:39 PM · 3 requests · claude-haiku-4-5-20251001 · $0.004 · 2 blocked (Customer ID, email_address, person_name)
Fix the failing invoice test for CUST-[BLOCKED:customer_id] in src/billing/invoice.ts.
Restore any of the last 10. Changes apply within a minute.
Give a reason, and within a minute every AI request is refused.
A private security team channel, plus a summary for each lead every working day.
They keep Claude Code, their own Claude plan, and the way they already work. Setup is one command, once.
curl -fsSL https://aidlc.moring.ai/onboard/install.sh | bashEveryone gets the same setup, so you can open each other’s projects and get the same results. Your own settings and notes stay.
Quit and reopen it, then check with /status. Use Claude Code in the terminal or VS Code: the Claude desktop app’s Code tab isn't covered.
Mac, Linux, WSL, or Windows, with no admin rights. The developer approves a short code in the browser and is done.
Works with Claude Code in both, on the developer’s own Claude plan.
If someone stops going through the gateway, they get a Slack message with the fix.
See every prompt, set the policy, and keep a record you can show.
Developers keep Claude Code. The gateway handles the checks.
For companies with customer data, audits such as SOC 2, or contracts that limit what goes to AI.
Claude Code sends the whole conversation with each request, including the files it read, command output, and even .env files with passwords and keys. So secrets can leave with ordinary prompts, and nobody notices. Moring AI checks every request before it leaves and replaces a secret with a label such as [REDACTED:aws_access_key_id], so Claude still understands the request without seeing the value.
Put one checkpoint between every developer’s Claude Code and Claude, apply one policy there, and keep a record. The Moring AI gateway runs in your own AWS account, so your data never passes through Moring AI, and it lets through only the models you approve. Start in monitor mode with one team, see what Claude Code sends, then switch to enforce.
Your provider bill doesn’t show which developer or session the spend came from. Moring AI logs the cost of every Claude Code request and shows each developer’s sessions, requests, models, and cost, so you see where the spend goes and spot unusual use early.
Claude Code keeps session history on each developer’s machine, which doesn’t help when an auditor or a customer asks what went out. Moring AI records every request after cleaning it, so you own the audit trail and it holds no secrets. The AI sessions view shows each conversation, what the policy did on each request, and whether a value came from the developer or from Claude Code itself.
Every check takes some time. In Moring AI’s load tests, the gateway added about 0.3 s to a typical request (a load-test figure, not a customer production figure). If a check can’t run, the gateway refuses the request, so nothing leaves unchecked.
No. Developers keep Claude Code, their own Claude plan, and the way they already work. Setup is one command on Mac, Linux, WSL, or Windows, with no admin rights, and it works in the terminal and in VS Code.
If someone stops going through the gateway, Moring AI sends them a Slack message with the fix, and your security team sees who isn’t on the gateway in the Security Console.
Start in monitor mode with one team. See what Claude Code sends, then switch to enforce when you are ready.