Fastly (BYOCDN)

This configuration routes agentic traffic (requests from AI bots and LLM user agents) to the Edge Optimize backend service (live.edgeoptimize.net). Human visitors and SEO bots continue to be served from your origin as usual. To test the configuration, after the setup is complete, look for the header x-edgeoptimize-request-id in the response.

Prerequisites

Before setting up the Fastly VCL rules, ensure you have:

Steps to retrieve your production Edge Optimize API key:

Additionally, if you require any help with the above steps, reach out to your Adobe account team or llmo-at-edge@adobe.com.

Use a staging hostname when you want to test Optimize at Edge in a lower environment before production traffic uses the routing rules.

Prerequisites

Steps

If you need help, contact llmo-at-edge@adobe.com.

Configuration

Add the following three VCL snippets to your Fastly service. These snippets handle routing agentic requests to Edge Optimize, cache-key separation, and failover to your default origin.

vcl_recv snippet

unset req.http.x-edgeoptimize-url;
unset req.http.x-edgeoptimize-config;
unset req.http.x-edgeoptimize-api-key;

if (!req.http.x-edgeoptimize-request
    && req.http.user-agent ~ "(?i)(AdobeEdgeOptimize-AI|ChatGPT-User|GPTBot|OAI-SearchBot|PerplexityBot|Perplexity-User)") {
  set req.http.x-forwarded-host = req.http.host; # required for identifying the original host
  set req.http.x-edgeoptimize-url = req.url; # required for identifying the original url
  set req.http.x-edgeoptimize-config = "LLMCLIENT=TRUE;"; # required for cache key
  set req.http.x-edgeoptimize-api-key = "<YOUR API KEY>"; # required for identifying the client
  set req.backend = F_EDGE_OPTIMIZE;
}

vcl_hash snippet

if (req.http.x-edgeoptimize-config) {
  set req.hash += "edge-optimize";
  set req.hash += req.http.x-edgeoptimize-config;
}

vcl_deliver snippet

if (req.http.x-edgeoptimize-config && resp.status >= 400) {
  set req.http.x-edgeoptimize-request = "failover";
  set req.backend = F_Default_Origin;
  restart;
}

if (!req.http.x-edgeoptimize-config && req.http.x-edgeoptimize-request == "failover") {
  set resp.http.x-edgeoptimize-fo = "1";
}

Failover

The vcl_deliver snippet handles failover automatically. If Edge Optimize returns a 4XX or 5XX error, the request is restarted and routed back to your default origin so the end-user still receives a response. Failover responses include the x-edgeoptimize-fo: 1 header.

Verify the setup

After completing the setup, verify that bot traffic is being routed to Edge Optimize and that human traffic remains unaffected.

1. Test bot traffic (should be optimized)

Simulate an AI bot request using an agentic user-agent:

curl -svo /dev/null https://www.example.com/page.html \
  --header "user-agent: chatgpt-user"

A successful response includes the x-edgeoptimize-request-id header, confirming that the request was routed through Edge Optimize:

< HTTP/2 200
< x-edgeoptimize-request-id: 50fce12d-0519-4fc6-af78-d928785c1b85

2. Test human traffic (should NOT be affected)

Simulate a regular human browser request:

curl -svo /dev/null https://www.example.com/page.html \
  --header "user-agent: Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_7) AppleWebKit/537.36"

The response should not contain the x-edgeoptimize-request-id header. The page content and response time should remain identical to before enabling Optimize at Edge.

3. How to differentiate between the two scenarios

4. Staging domain (optional)

If you use a staging hostname and staging API key from LLM Optimizer, add the same VCL snippets to your staging Fastly service using the staging API key. Then verify bot traffic on the staging host:

curl -svo /dev/null https://staging.example.com/page.html \
  --header "user-agent: chatgpt-user"

Replace https://staging.example.com/page.html with your real staging URL and path. A successful response includes the x-edgeoptimize-request-id header.

The status of the traffic routing can also be checked in the LLM Optimizer UI. Navigate to Customer configuration and select the CDN configuration tab.

To learn more about Optimize at Edge, including available opportunities, auto-optimization workflows, and FAQs, return to the Optimize at Edge overview.