\u2013 The firewall runs entirely client-side, so no data ever leaves your environment.
\u2013 It focuses on *retrieval-time* risks, not output moderation \u2014 so the LLM never sees poisoned chunks in the first place.
\u2013 Policies are YAML: you can choose to deny, allow, or just re-rank risky docs (based on recency, provenance, relevance).
\u2013 Overhead is low: scanners are regex/heuristic, so for ~5\u201320 chunks it adds only a few ms.
I\u2019d love feedback on two things in particular:
1. Do you think retrieval-time filtering belongs in the pipeline, or should it all be done at ingest/output?
2. If you\u2019ve got prompt injection payloads or edge cases you use to test your own RAG stacks, I\u2019d love to try them against this.
Thanks for taking a look \u2014 always happy to hear critique, especially from folks running LangChain/LlamaIndex in production.","title":null,"type":"comment","url":null},{"author":"talbuilds","children":[],"created_at":"2025-08-30T15:07:00.000Z","created_at_i":1756566420,"id":45075303,"options":[],"parent_id":45068582,"points":null,"story_id":45068582,"text":"300+ installs in 24h, RAG Firewall now with GraphRAG support.","title":null,"type":"comment","url":null},{"author":"danshalev7","children":[],"created_at":"2025-09-01T07:02:44.000Z","created_at_i":1756710164,"id":45090229,"options":[],"parent_id":45068582,"points":null,"story_id":45068582,"text":"Very interesting","title":null,"type":"comment","url":null}],"created_at":"2025-08-29T19:48:12.000Z","created_at_i":1756496892,"id":45068582,"options":[],"parent_id":null,"points":1,"story_id":45068582,"text":"RAG pipelines are great, but they can still retrieve "toxic" chunks:\n\u2013 prompt injection attempts\n\u2013 leaked API keys/secrets\n\u2013 stale or conflicting content\n\u2013 unapproved external URLs
We built an open-source "retrieval firewall" that scans chunks before they reach the LLM:\n\u2013 denies injection & secrets\n\u2013 flags/reranks PII, encoded blobs, untrusted URLs\n\u2013 audit log (JSONL) of all decisions\n\u2013 drop-in wrappers for LangChain and LlamaIndex retrievers
Install: pip install rag-firewall\nRepo: https://github.com/taladari/rag-firewall
Curious if others here handle retrieval-time risks, or just ingest/output filtering.\nWould love feedback and red-team payloads.","title":"Show HN: RAG Firewall \u2013 retrieval-time guardrails for LangChain/LlamaIndex","type":"story","url":"https://github.com/taladari/rag-firewall"}