built for teams running coding agents

Enterprise security
for your coding agents.

Rechazo AI is coding agent security that sits between your engineers and the agents they work with: Claude Code, Codex, Gemini CLI, Cursor and Copilot. It inspects every prompt, watches every action an agent takes, and keeps credentials out of the conversation, so your whole company can use AI agents safely.

Six ways AI agents put your company at risk

What can go wrong when AI agents work inside your company, and how Rechazo handles each one.

Passwords end up in AI chats

Engineers paste error messages to get help. Those messages often contain company passwords, which are then stored by an outside AI company.

Passwords are removed before anything is sent.

Customer data leaves the company

Names, emails and account details get copied into prompts while fixing a bug. Nobody notices, and nobody can take it back.

Personal data is stripped out automatically.

Agents act without asking

An agent can delete files, change live servers or publish code on its own. One wrong step and customers feel it.

Risky actions wait for a person to approve them.

Agents can be tricked

A web page or file can hide instructions that tell the agent to do something harmful. The engineer never sees them.

Every action is checked, whatever told the agent to do it.

No record of who did what

When something breaks, there's no trail of which agent did it, for which person, or why. Audits become guesswork.

Every prompt and action is logged with a name.

Banning AI pushes it out of sight

Block the tools and engineers switch to personal accounts on their own laptops, where you can see nothing at all.

Teams keep their agents, on company terms.

Your AI agent has the same access as your engineer

When an engineer uses an AI agent, it logs in as them: cloud, code, payments, customer data. If the agent leaks a password or gets tricked, all of it is exposed.

AI agent
1
Systems it can open
0
Passwords & keys
0

Each line is a door the agent can open, with nobody checking.

Rechazo stands between the agent and every system, and checks each action first.

AWS
GitHub
Stripe
PostgreSQL
Slack
Customer data
Kubernetes
Vercel
Google Cloud
Sentry
OpenAI
Terraform
npm
MongoDB

Open access: nothing checks what the agent does

Your engineer's AI agent

Rechazo

Protection across prompts, actions and credentials

Rechazo runs on every engineer's laptop, right beside their AI agent. It covers five areas without changing how anyone works.

  1. Rechazo · Protection on the laptop

    Engineer's laptopinside your limits

    • Edit the login pagesrc/login.tsx allowed
    • Run the testsnpm test allowed
    • Delete the project folderrm -rf ~/project blocked
    • Send files to an unknown sitecurl -T .env paste.io blocked

How Rechazo goes live across a team in an afternoon.

No change to how engineers work. They keep the agent they like; Rechazo runs alongside it on every laptop.

  1. 1

    Install on every laptop

    • Pushed with the device management you already use
    • Runs quietly next to each engineer's agent
    • Native on macOS, Linux and Windows
    via your device management
    laptops protected0 / 48

    Engineers keep their agent. Nothing new to learn.

  2. 2

    Choose your rules

    • Start from sensible defaults
    • Switch any rule on or off in a click
    • Changes reach every laptop in seconds
  3. 3

    See everything in one place

    • Security sees what was stopped, and why
    • Engineering sees how agents are used across the team
    • Unusual activity is flagged and the seat paused
    console.rechazo.si

    Prompts checked

    0

    Keys removed

    0

    Customer data removed

    0

    Actions stopped

    0

    Prompts checked per day · 30 days

    • 03:12Login from a new deviceSeat paused
    • 02:48API key removed from a promptRemoved
    • 02:31Customer email removedRemoved
    • 01:57Terraform apply needs sign-offApproval
    • 01:20Unapproved tool serverBlocked

Works with the agents your team already uses

  • Claude Code
  • Codex
  • Gemini CLI
  • Cursor
  • Copilot

Checked in memory

Prompts are checked as they pass through. Keys are removed before anything is saved.

Decisions, not conversations

By default we keep what was decided, not what was said. You choose if, and how long, to keep more.

Never used for training

Your prompts, code and keys are never used to train any model.

Join the waitlist

Credential protection for every system your agents can reach

  • Claude Code
  • AWS
  • Supabase
  • PostgreSQL
  • Stripe
  • GitHub
  • Google Cloud
  • Azure
  • MongoDB
  • Redis
  • Vercel
  • Cloudflare
  • OpenAI
  • Kubernetes
  • Terraform
  • HashiCorp Vault
  • Snowflake
  • 1Password
  • Firebase
  • MySQL
  • Claude Code
  • AWS
  • Supabase
  • PostgreSQL
  • Stripe
  • GitHub
  • Google Cloud
  • Azure
  • MongoDB
  • Redis
  • Vercel
  • Cloudflare
  • OpenAI
  • Kubernetes
  • Terraform
  • HashiCorp Vault
  • Snowflake
  • 1Password
  • Firebase
  • MySQL
  • Slack
  • Datadog
  • Sentry
  • npm
  • Docker
  • GitLab
  • Okta
  • Auth0
  • Jira
  • Linear
  • Notion
  • PlanetScale
  • Neon
  • Upstash
  • Prisma
  • Clerk
  • Resend
  • DigitalOcean
  • Netlify
  • Elasticsearch
  • Kafka
  • Grafana
  • PagerDuty
  • Shopify
  • Slack
  • Datadog
  • Sentry
  • npm
  • Docker
  • GitLab
  • Okta
  • Auth0
  • Jira
  • Linear
  • Notion
  • PlanetScale
  • Neon
  • Upstash
  • Prisma
  • Clerk
  • Resend
  • DigitalOcean
  • Netlify
  • Elasticsearch
  • Kafka
  • Grafana
  • PagerDuty
  • Shopify

Database URLs, cloud credentials, API tokens and webhooks. Rechazo keeps them out of prompts and scoped to the commands that need them, across every system your AI agents touch. How to keep API keys out of AI coding agents

AI agent incidents are already happening

Leaked keys and agent mistakes are on the public record, and the numbers climb every year.

0

secrets found in public AI tool configuration files. 2,117 still worked.

GitGuardian

0

security flaws filed against Claude Code alone since June 2025.

NIST NVD

0–90%

of a state-sponsored espionage campaign was run through an AI coding agent.

Anthropic

The public record

Credential leaks are compounding.

Not a projection: two primary-source series, secrets in public commits and CVEs filed against Claude Code.

Public GitHub · new secrets detected per year

28.65M

detected in 2025, up 34% on 2024.

~2×

leak rate of Claude Code co-authored commits vs GitHub baseline.

+81%

growth in leaked AI-service credentials.

24,008

secrets in public MCP configuration files.

GitGuardian, State of Secrets Sprawl 2026Y-axis starts at zero. 2021–2024 are the report's rounded chart values; 2025 is exact.

From the library

Field guides for securing AI coding agents

All guides
keep-secrets-out-of-ai-coding-agents.mdGuide

How to keep API keys and .env files away from AI coding agents

26 October 20266 min readGuide

FAQ

AI coding agent security, answered.

What is Rechazo AI?

Rechazo AI (rechazo.si, also written Rechazo) is an AI coding agent security company. The name is Spanish for "rejection": Rechazo rejects the small fraction of prompts and agent actions that should not happen, so the rest can happen freely. It inspects every prompt before it leaves the company, stops risky agent actions on the machine and keeps credentials out of the conversation.

What is AI coding agent security?

AI coding agent security is the set of controls that keep tools like Claude Code, Codex, Gemini CLI, Cursor and Copilot from leaking secrets, source code or customer data, and from taking risky actions on a developer machine. It covers what goes into prompts, what the agent is allowed to run, and which tools and MCP servers it can reach.

Read the AI coding agent security guide

How does Rechazo keep secrets out of prompts?

Rechazo sits between your engineers and their agents and inspects each prompt before it leaves the company. Credentials, keys and sensitive values are caught in the prompt, so they can be redacted or the prompt blocked, rather than being sent to a model provider.

Which AI coding agents does Rechazo cover?

Rechazo works with all the major AI coding agents, including Claude Code, Codex, Gemini CLI, Cursor and Copilot, and teams can run several side by side.

What is agent detection and response (ADR)?

Agent detection and response watches what an AI agent actually does on the machine, such as commands it runs, files it reads, and MCP servers or skills it loads, and can stop a risky action before it runs. It complements prompt inspection, which only sees what is sent to the model.

Read the AI agent runtime security guide

Can Rechazo stop prompt injection?

No tool can promise to eliminate prompt injection. Rechazo limits the damage: it keeps credentials out of the conversation and enforces policy on the actions an agent is allowed to take, so a hijacked agent has less to steal and less it can do.

Read the Prompt injection guide

Does Rechazo replace Claude Code's built-in security controls?

No. Rechazo works with the controls the agent already ships with, such as permissions and managed settings, and adds inspection, enforcement and an audit trail across the whole company.

Read the Claude Code security guide

Is Claude Code safe for enterprise use?

It can be, with the right configuration. Claude Code enforces permission rules in the tool rather than the model, lets administrators push managed settings that developers cannot override, can sandbox shell commands at the operating-system level and can restrict MCP servers to an allowlist. None of those inspects what a developer pastes into a prompt, and each covers Claude Code only.

Read the Claude Code security guide

How do I stop AI coding agents from reading .env files and API keys?

Deny the agent access to secret files in each tool's settings, such as Read deny rules in Claude Code, remove secrets from the environment that agent commands inherit, and give agents short-lived, narrowly scoped tokens. Deny rules alone are not enough, because shell commands and scripts can still reach files and environment variables.

Read the guide to keeping secrets out of AI coding agents

What is MCP security?

MCP security is deciding which Model Context Protocol servers a coding agent may connect to, what each one can reach, and how you notice when one changes after approval. The main risks are instructions hidden in tool descriptions, definitions that change after review, over-scoped tokens and local servers running with the developer's own privileges.

Read the MCP security guide

How do I get access to Rechazo AI?

Rechazo AI is deployed with each customer rather than downloaded. Join the waitlist and the team will schedule a walkthrough on a real coding agent workflow.