AgentVault LLM Integration

Instructions

• •

  OpenRouter setup

  • •Read the OpenRouter API key from an environment variable (e.g., OPENROUTER_API_KEY).
  • •Use requests or a small HTTP client to call the OpenRouter API.
  • •Configure sensible timeouts and minimal retries to avoid hanging processing jobs.
• •

  Model selection

  • •Default to high-quality models such as anthropic/claude-3.5-sonnet for rewriting and structuring notes.
  • •Optionally, provide a way to switch to cheaper models (e.g., strong open-source models served via OpenRouter) when cost is a concern.
• •

  Prompt design

  • •Prompts should instruct the model to:
    • •Preserve original meaning and nuance.
    • •Keep code blocks and links intact.
    • •Enhance clarity and structure for downstream AI agents.
  • •Include explicit instructions not to add fabricated details or hallucinate content.
• •

  Usage pattern

  • •For each note body:
    • •Build a prompt that includes relevant context (frontmatter, headings) without exceeding token limits.
    • •Send a single request per note, or small batches, to OpenRouter.
  • •Use async concurrency where appropriate, but respect rate limits and error responses.
• •

  Safety and privacy

  • •Never log raw prompts or responses that contain user note content.
  • •Log only high-level metadata, such as:
    • •Number of tokens used.
    • •Model names.
    • •Success/failure statuses.

Examples

• •Given a markdown note body and frontmatter, outline:
  • •How to build a prompt string tailored to AgentVault’s use case.
  • •How to construct the OpenRouter API request.
  • •How to incorporate the LLM’s response back into the final markdown file.