# TryBrake

> TryBrake is a safety layer and kill-switch for AI agents and automations. Before an agent refunds, emails, deletes or pays, it asks TryBrake, which applies the business's rules: let it through, block it, or hold it for a human to approve. Every action is logged and can be replayed, and every agent can be stopped in one click.

TryBrake works with no-code automations (n8n, Make, Zapier), AI coding assistants (Claude Code, Codex, Cursor, Gemini CLI) and custom agents (Python, JavaScript or any HTTP client). Rules are enforced by a deterministic engine, without AI at decision time, so an agent can't talk its way past them. AI is only used to suggest rules, which the customer reviews. Data is hosted in the EU (Paris).

## Key facts

- Decisions per action: allow, block, or require human approval (approve or deny from the dashboard)
- Rules on action details (e.g. refund amount over 50 €), time of day, weekends, and repeats (e.g. "never twice for the same order")
- Volume limits (e.g. at most 30 emails per hour) to stop runaway loops
- Global kill-switch, per-agent pause, per-action off switch
- Run replay with the agent's reasoning and each action's result; audit log and printable audit report
- n8n and Make: connect the account and TryBrake adds itself to the workflow, reversibly; it analyses the workflow and proposes rules for each action
- Coding assistants: one text to paste into the assistant, which installs the guard itself
- Connectors fail closed: if TryBrake can't be reached, the action does not run

## Pricing

14-day free trial with Pro features, no card needed. Monthly plans by number of agents:
- 1–10 agents: Essential 49 €/month, Pro 99 €/month
- 11–25 agents: Essential 99 €/month, Pro 199 €/month
- 25+ agents: Essential 199 €/month, Pro 399 €/month
- Custom: more agents, custom terms
Pro adds alerts, volume limits and 1-year history (Essential keeps 30 days).

## Pages

- [Home](https://trybrake.com/): what TryBrake does, how it works, pricing
- [Free website check](https://trybrake.com/#main): what an unguarded AI agent could break in a business, with a risk score
- [Pricing](https://trybrake.com/#pricing)
- [Integrations](https://trybrake.com/integrations): n8n, Make, Claude Code, ChatGPT & OpenAI (Codex), Zapier, Cursor, Gemini CLI, Python, JavaScript
- [Security](https://trybrake.com/security): how decisions, keys, data and approvals are protected
- [Docs](https://trybrake.com/docs): the check API, approvals, connectors and rules
- [Sign up](https://trybrake.com/login?mode=signup)

## Legal

- [Legal notice](https://trybrake.com/legal)
- [Privacy policy](https://trybrake.com/privacy)
- [Terms of service](https://trybrake.com/terms)
- [Cookie policy](https://trybrake.com/cookies)

Published by Elie Amar (France).