BLAZLE is an in-flight security layer for AI agents. It inspects every prompt, shell command and code diff before it leaves your environment.

Problem it solves: AI agents (Cursor, Claude, Copilot, custom pipelines) can paste API keys into prompts, run destructive commands, or follow prompt injections hidden in content they read. Most tools only detect this after the fact. BLAZLE stops it at the moment the request is produced.

Key features:

• Blocks or redacts secrets (AWS, GitHub, Stripe, private keys…) and personal data before they reach the model or repo.
• Anti-obfuscation: every payload is normalized before evaluation, so secrets hidden with encoding (base64, hex, URL), look-alike characters, invisible characters or token splitting are caught like plain ones.
• An AI layer blocks prompt injection. Destructive commands (rm -rf, privilege escalation) are blocked too.
• The GitHub App scans every pull request. With Hard Block on, the Merge button stays locked until the issue is fixed.
• MCP Copilot lets you manage security in plain language. Every change waits for human approval, and destructive actions need a second approver.
• A tamper-evident audit log, with SOC 2 and MAS-TRM evidence export.

Who it is for: dev and platform teams running AI agents in production, and the security and compliance teams who need control over what those agents send out.

How to get started: sign up free at blazle.io (5,000 requests/month, no credit card), create an API key, then integrate via:

• REST API
• Local terminal proxy
• VS Code/Cursor extension
• Git pre-commit hook
• GitHub App
• MCP (Cursor, Claude)

A setup video under 3 minutes is at blazle.io/#setup. Self-hosting is available for enterprises.

What makes it different:

• Interception while the request is in flight, not scanning after the fact.
• Rule checks take about 0.2 ms; the AI check takes 130-150 ms and only runs when needed (measured in production).
• No payload is sent to third-party AI services.
• Human approval for agent actions.
• The team uses BLAZLE to protect all of its own repositories.