Skip to main content

Amazon Bedrock (AI)

AI features use Amazon Bedrock with Claude models (often via inference profile IDs in us. regions).

Configuration​

  • Environment variables on Lambda (e.g. BEDROCK_MODEL_ID) and IAM permissions for bedrock:InvokeModel.
  • Prompts are maintained in backend/lambda-functions/bedrock_prompts.py and wired in edr-api-handler.py.
  • See repository backend/BEDROCK_SETUP.md, BEDROCK_MODEL_ACCESS.md, and AI_PROMPTS.md.

Modes​

  • Standard analysis — JSON output (summary, classification, key indicators, …).
  • RAG analysis — When Pinecone MITRE index is configured, retrieval augments the prompt (SYSTEM_PROMPT_AI_ANALYSIS_RAG).

Follow-up chat​

  • General questions use SYSTEM_PROMPT_AI_CHAT.
  • Short verdict-style questions use SYSTEM_PROMPT_AI_CHAT_VERDICT for a compact, consistent layout.

Redeploy Lambda after changing prompts: backend/deploy-lambda-with-rag.ps1.