BLAZLE
In-flight security for AI agents: stop prompt injection and leaks before egress. BLAZLE inspects every prompt, shell command, code diff and MCP tool call before it leaves your environment.
AI use BLAZLE is a self-hosted AI security proxy that sits between AI agents and the world. Agents send prompts, shell commands, code diffs and MCP tool calls through BLAZLE; an own, self-hosted detection model inspects each one before egress to redact secrets, block prompt injections in 30+ languages and hold risky actions for human approval. Through the BLAZLE MCP Gateway and MCP Copilot, agents also call per-tool approve/block controls and get Copilot assistant answers that use OpenAI only as an optional downstream provider.
BLAZLE sits between AI agents and the world. Every prompt, shell command, code diff and MCP tool call is inspected before it leaves: secrets redacted, prompt injections blocked in 30+ languages, risky actions held for human approval. Works as an edge proxy for VS Code, Cursor, Claude Code and Antigravity, a REST API, an MCP Gateway with per-tool approve/block, and a GitHub App that scans PRs before merge. Detection is self-hosted, so payloads never go to a third-party AI.
- Category
- Security & Privacy
- Agent access
- REST API
- AI models
- Own model
- Trains on customer data
- No
- HQ
- United States of America
- Pricing
- Freemium, Free forever tier: 5,000 intercepted requests/month, no credit card required. Paid Pro and Business plans available.
The vendor ships an MCP server. Endpoint and authentication as stated by the vendor; whether I tested it myself is in the review.
- Authentication
- No auth