Routiium launches self-hosted LLM gateway with built-in safety guards — blocks prompt injections and unsafe tool use without code changes

Routiium (open-source project) released a self-hosted gateway that sits between your app and AI providers (OpenAI, vLLM, Ollama, AWS Bedrock). It adds API key management, request filtering, and safety checks while staying compatible with existing OpenAI SDKs — developers point their code at Routiium's local port 8088 instead of the provider, no SDK rewrite needed.

The core safety feature: a built-in "judge" that automatically detects and blocks prompt injections, data exfiltration attempts, and unsafe tool calls before they reach the AI model. It uses both deterministic rules (pattern matching) and optional LLM-based judgment, then either rejects the request or downgrades it to a safer AI model alias (auto, fast, balanced, safe, secure, premium) based on risk level — agentic apps get a denial response and can retry safely instead of crashing.

For teams shipping LLM features in production, this eliminates the need to build custom safety layers or trust providers' guardrails alone. Engineering teams can install via npm, run `routiium serve`, and gain rate limiting, analytics, managed API keys, and policy routing out of the box. Non-technical users benefit because apps using Routiium will reject jailbreak attempts and dangerous requests automatically.

Routiium is open-source and available via npm (`npm install -g routiium`), Docker, or cargo. Teams can run it self-hosted on their own infrastructure or connect it to a remote Router-compatible policy service for centralized governance across multiple deployments.