RAG Knowledge Base Poisoning Sentinel
Review newly ingested knowledge chunks for instruction injection, provenance gaps, and suspicious content before retrieval.
The full canvas, before you import it
Click any node to see its config.
Click a node to select it — same as the Heym editor; the panel shows its settings.
9 nodes · Free & source-available
RAG Knowledge Base Poisoning Sentinel
Add a quarantine layer between document ingestion and production retrieval. This workflow reviews pending knowledge chunks for hidden instructions, source impersonation, encoded payloads, unrelated redirects, and missing provenance before they can influence an agent.
What this workflow does
- RagSafetySchedule runs every 15 minutes
- FindPendingKnowledgeChunks loads unreviewed chunks from DataTable
- IterateKnowledgeChunks processes each chunk independently
- InspectKnowledgeChunk assigns quarantine, review, or accept status
- RouteKnowledgeSafety sends each decision to a different path
- Quarantined chunks are marked and removed from the publish queue
- Ambiguous chunks alert the knowledge security team
- Accepted chunks are recorded as approved
Use cases
- RAG poisoning detection
- Indirect prompt injection screening
- Knowledge base ingestion quarantine
- Retrieval provenance enforcement
Setup
Select the DataTable that receives pending knowledge chunks and map sourceUrl, contentText, contentHash, and ingestionStatus fields. Connect an LLM credential and Slack. Keep deterministic file validation, access controls, and source allowlists around this semantic review.
How to import this template
- 1Click Import → Copy JSON on this page.
- 2Open your Heym and navigate to a workflow canvas.
- 3PressCmd+V/Ctrl+V— nodes appear instantly.
- 4Add your API keys in the node config panels and click Run.
Discover more automations
- AI AgentsBatch LLM Status TrackerSend an array through the OpenAI Batch API, branch on live status updates, and collect the final per-item results.
- AI AgentsBuild Your First AI AgentA beginner-friendly interactive AI agent with conversation memory — type a message and get context-aware replies.
- AI AgentsBlind Eval TrioThree cross-lab agents evaluate any decision blind: steelman defends, stress_test attacks, gap_finder finds what's missing. No synthesizer — you integrate.
- AI AgentsAI Research Brief Supabase ArchiveTurn research notes into a structured AI brief and save the finished Markdown in Supabase.
- AI AgentsAnime Multi-Source Research BriefBuild a spoiler-aware anime research brief from Jikan profile, character, and review endpoints with deliberate rate-limit waits.
- Dev & IT OpsHTML Status PageCheck an upstream service and answer a browser GET with a rendered status page instead of JSON.
- Dev & IT OpsWorkflow Change Audit LogCapture every workflow create, update, and delete on your Heym instance and post a batched summary to Slack.
- Customer SupportHITL Support Reply AgentDraft a customer-facing support response, pause for human approval, then continue with the reviewed reply.
- Dev & IT OpsCodex PR Fix AgentSend a coding task to Codex, open a draft PR when it succeeds, and notify Slack when Codex needs follow-up input.
- Dev & IT OpsOpenCode Go PR Fix AgentSend a coding task to OpenCode Go, open a review-ready pull request, and return the implementation result.
- Dev & IT OpsAgent-Guided Codex PR DispatcherLet an Agent inspect a GitHub request, prepare a confirmed Codex task, open a ready PR, and notify Slack when Codex needs input.
- Marketing & SEOReddit Subreddit GET Tool CallLet an Agent call a no-key HTTP GET tool that fetches hot posts from a subreddit JSON feed.