ArcFlow

Origin and rate limiting

Relay enforces two site-level controls before proxying to arcflow-server: Origin allowlist and requests per minute. Both are configured when a site is created or patched via admin API, or inline in ARCFLOW_RELAY_SITES_JSON for BYO Relay.

allowed_origins

Each site stores allowed_origins as a string array (e.g. ["https://www.acme.com", "https://app.acme.com"]).

On each browser request, Relay reads the Origin header (or Referer fallback in some paths) and compares against the list. Mismatch returns 403:

json
{
 "error": {
 "code": "origin_not_allowed",
 "message": "[ArcFlow Relay] Origin not allowed for this site."
 }
}

Setting origins at create

bash
curl -s -X POST http://localhost:8080/v1/admin/sites \
 -H "Authorization: Bearer dev-admin" \
 -H "Content-Type: application/json" \
 -d '{
 "display_name": "Acme",
 "allowed_origins": ["https://www.acme.com"],
 "rate_limit_rpm": 60,
 "allow_inline": false
 }'

Updating origins

bash
curl -s -X PATCH "http://localhost:8080/v1/admin/sites/site_abc123" \
 -H "Authorization: Bearer dev-admin" \
 -H "Content-Type: application/json" \
 -d '{"allowed_origins":["https://www.acme.com","https://staging.acme.com"]}'

Redeploy or reload Relay if using JSON file config so in-memory site store picks up changes.

Development vs production

EnvironmentTypical Origin
Vite dev serverhttp://localhost:5173
Preview deployhttps://preview-xyz.vercel.app
ProductionExact production origin, no wildcard

Relay does not support * origin in production configurations. Add each deploy preview origin explicitly or use separate dev sites.

Local test with curl

Relay expects a browser-like Origin:

bash
curl -s -X POST "http://localhost:8090/v1/sites/s_dev/runs" \
 -H "Authorization: Bearer st_live_devtoken" \
 -H "Origin: http://localhost:5173" \
 -H "Content-Type: application/json" \
 -d '{"workflow_ref":{"name":"chat","version":"^1.0.0"},"input":"test"}'

Wrong origin:

bash
curl -s -o /dev/null -w "%{http_code}\n" \
 -X POST "http://localhost:8090/v1/sites/s_dev/runs" \
 -H "Authorization: Bearer st_live_devtoken" \
 -H "Origin: https://evil.example" \
 -H "Content-Type: application/json" \
 -d '{"workflow_ref":{"name":"chat","version":"^1.0.0"},"input":"test"}'

Expect 403.

rate_limit_rpm

rate_limit_rpm caps create + poll traffic per site per minute (implementation uses in-memory token bucket keyed by site id). Default 60 at site create.

When exceeded, Relay returns 429:

json
{
 "error": {
 "code": "rate_limited",
 "message": "[ArcFlow Relay] Rate limit exceeded."
 }
}

Tuning guidance

Site typeSuggested RPM
Internal demo60
Customer support widget120 to 300
High-traffic marketing pageScale Relay replicas; raise RPM with monitoring

Rate limits protect upstream LLM cost and server capacity. They do not replace WAF or bot management at the CDN edge.

Security implications

  • Wrong Origin config is the most common production misconfiguration. Symptom: browser CORS-like failures with 403 from Relay.
  • Leaked site token plus correct Origin still allows abuse within RPM. Rotate token immediately via admin rotate endpoint.
  • Site tokens are not LLM keys, but they still authorize workflow runs billed to your infrastructure.

Related pages

Previous

Request path

Next

BYO relay deployment