See llms.txt for all machine-readable content.

Back to Templates

Probe AISIX model aliases and generate health reports for SRE teams

Created by

Created by: Yilia Lin || yilialinn
Yilia Lin

Last update

Last update 2 days ago

Categories

Share

Quick Overview

This workflow runs on demand to list model aliases from an AISIX AI Gateway and sends a small OpenAI-compatible Chat Completions probe to each alias, then returns a consolidated health report that flags rate limits, auth errors, policy blocks, and other unhealthy responses.

How it works

  1. Runs when triggered manually.
  2. Sets the AISIX gateway base URL, probe prompt, and a maximum number of model aliases to test.
  3. Calls the AISIX /v1/models endpoint to fetch the model aliases visible to the configured caller API key.
  4. Builds a capped queue of aliases to probe, or returns an attention result if model discovery fails or no aliases are available.
  5. Sends one non-streaming Chat Completions request to AISIX /v1/chat/completions for each alias.
  6. Classifies each probe response as healthy, rate limited, policy blocked, auth failed, or unhealthy and captures latency plus relevant response headers.
  7. Aggregates results into an overall health report and outputs it on either a healthy or attention path.

Setup

  1. Create an n8n HTTP Header Auth credential that sets Authorization: Bearer <your AISIX caller API key> and select it in both AISIX HTTP Request steps.
  2. Update aisix_base_url, probe_prompt, and max_models in the configuration step to match your AISIX gateway and desired probe behavior.
  3. Ensure your AISIX instance supports OpenAI-compatible /v1/models and /v1/chat/completions endpoints and that the caller key can list and call at least one model alias.