Add a security checkpoint between an internal app and an LLM provider like OpenAI or Anthropic.
Run an AI firewall entirely inside a closed network with no outside internet access.
Enforce rate limits, access control, and content scanning on every LLM request and response.
Monitor and configure LLM traffic policies through a real-time management console.
| c6web/guardrails | 0labs-in/vision-link | alfons-fhl/cursor-plan2api | |
|---|---|---|---|
| Stars | 4 | 4 | 4 |
| Language | TypeScript | TypeScript | TypeScript |
| Setup difficulty | moderate | moderate | moderate |
| Complexity | 4/5 | 3/5 | 3/5 |
| Audience | ops devops | developer | developer |
Figures from each repo's GitHub metadata at analysis time.
The quick-start script provisions PostgreSQL and starts all services, but production use needs a separate hardening pass.
C6 Guardrails is an open source firewall that sits between an application talking to a large language model, like ChatGPT or Claude, and the AI provider it is calling. Every request going out and every response coming back passes through the firewall first, so it can check them against a set of security threat policies based on the OWASP list for LLM applications published for 2025. It works with any model, whether it runs locally on your own machine or through a cloud provider, and it can be deployed entirely inside a closed network with no outside internet access. At its core is an 8 stage pipeline written in Rust that each request and response moves through: authentication, rate limiting, access control, classification, enforcement, checks on any tools the AI is allowed to call, scanning of the response content, and a final content quality check. Alongside this there is a real time management console with 47 pages for watching and configuring the system. The project already supports connecting to OpenAI, Anthropic, Gemini, Ollama, and OpenRouter as the upstream providers being protected. The stated goal is to save every company from building this kind of AI security and compliance layer from scratch, since governance requirements increasingly demand something like this be in place. It was built by C6Web and Victor Tong as an open source platform meant to cover these needs out of the box rather than as a one off internal tool. Getting a demo running is a single script that provisions a PostgreSQL database, runs migrations and seed data, and starts every service at once, with the gateway on one local port and the management frontend on another. The README is explicit that this all in one demo build is meant for evaluation and development only, and points to a separate hardening guide for anything running in production. The system architecture separates the Rust based gateway engine, an Express backend, and a PostgreSQL database holding data, users, and logs, plus a Discord community for support. It ships under the Apache 2.0 license, with an added requirement to keep certain branding attribution intact, and comes with a default admin login you are prompted to change on first use.
An open source AI firewall that inspects traffic between applications and LLM providers like OpenAI or Anthropic against OWASP security threat policies.
Mainly TypeScript. The stack also includes Rust, TypeScript, PostgreSQL.
Free to use, modify, and distribute, including commercially, under Apache 2.0, but you must keep the required branding attribution.
Setup difficulty is rated moderate, with roughly 30min to a first successful run.
Mainly ops devops.
This repo across BitVibe Labs
double-check against the repo, no cap.