Client-side retrieval firewall for RAG systems — blocks prompt injection and secret leaks, re-ranks stale or untrusted content, and keeps all data inside your environment.

13 stars 1 forks 13 watchers Python Apache License 2.0
ai-security ai-security-tool compliance guardrails langchain llamaindex llamaindex-rag llm llm-compliance llm-security llms prompt-injection python rag rag-compliance retrieval-augmentation-generation retrieval-augmented retrieval-augmented-generation retrieval-augmented-generation-rag secrets-detection
3 Open Issues Need Help Last updated: Sep 4, 2025

Open Issues Need Help

View All on GitHub

AI Summary: This issue proposes adding a new test to verify the correctness of audit logs. The test will run a mini pipeline using a `DummyRetriever` with mixed document types and then assert that the generated JSONL audit log contains expected events like deny, allow, and rerank, effectively acting as a golden snapshot test to prevent regressions.

Complexity: 3/5
good first issue testing

Client-side retrieval firewall for RAG systems — blocks prompt injection and secret leaks, re-ranks stale or untrusted content, and keeps all data inside your environment.

Python
#ai-security#ai-security-tool#compliance#guardrails#langchain#llamaindex#llamaindex-rag#llm#llm-compliance#llm-security#llms#prompt-injection#python#rag#rag-compliance#retrieval-augmentation-generation#retrieval-augmented#retrieval-augmented-generation#retrieval-augmented-generation-rag#secrets-detection

AI Summary: This issue requests the addition of a new runnable example demonstrating the integration of the Chroma vector store with a LangChain retriever. The goal is to provide a practical RAG application example to aid user adoption, with acceptance criteria including a new Python file and a README update.

Complexity: 1/5
documentation good first issue

Client-side retrieval firewall for RAG systems — blocks prompt injection and secret leaks, re-ranks stale or untrusted content, and keeps all data inside your environment.

Python
#ai-security#ai-security-tool#compliance#guardrails#langchain#llamaindex#llamaindex-rag#llm#llm-compliance#llm-security#llms#prompt-injection#python#rag#rag-compliance#retrieval-augmentation-generation#retrieval-augmented#retrieval-augmented-generation#retrieval-augmented-generation-rag#secrets-detection

AI Summary: This issue proposes adding a new `SQLInjectionScanner` to a RAG firewall. The scanner will use regex patterns to detect common SQL injection attempts (e.g., "UNION SELECT", "DROP TABLE") in retrieved text, preventing malicious queries from exploiting surfaced database documentation. It requires a new scanner file, pattern matching, and unit tests.

Complexity: 2/5
enhancement good first issue

Client-side retrieval firewall for RAG systems — blocks prompt injection and secret leaks, re-ranks stale or untrusted content, and keeps all data inside your environment.

Python
#ai-security#ai-security-tool#compliance#guardrails#langchain#llamaindex#llamaindex-rag#llm#llm-compliance#llm-security#llms#prompt-injection#python#rag#rag-compliance#retrieval-augmentation-generation#retrieval-augmented#retrieval-augmented-generation#retrieval-augmented-generation-rag#secrets-detection