diff --git a/api/changelog.d/gunicorn-max-requests.fixed.md b/api/changelog.d/gunicorn-max-requests.fixed.md
new file mode 100644
index 0000000000..e43a4ea57a
--- /dev/null
+++ b/api/changelog.d/gunicorn-max-requests.fixed.md
@@ -0,0 +1 @@
+Gunicorn workers are now recycled after 10000 requests (plus up to 1000 of jitter) to cap memory growth, configurable with `GUNICORN_MAX_REQUESTS` and `GUNICORN_MAX_REQUESTS_JITTER`
diff --git a/api/src/backend/config/guniconf.py b/api/src/backend/config/guniconf.py
index 8b17ad669d..5880d9f45c 100644
--- a/api/src/backend/config/guniconf.py
+++ b/api/src/backend/config/guniconf.py
@@ -53,6 +53,10 @@ preload_app = not DEBUG
 # that may take longer, such as complex API operations.
 timeout = env.int("GUNICORN_TIMEOUT", default=120)
 
+# Recycle workers after N requests (+ jitter) to cap memory growth. GUNICORN_MAX_REQUESTS=0 disables.
+max_requests = env.int("GUNICORN_MAX_REQUESTS", default=10000)
+max_requests_jitter = env.int("GUNICORN_MAX_REQUESTS_JITTER", default=1000)
+
 # Logging
 logconfig_dict = DJANGO_LOGGERS
 gunicorn_logger = logging.getLogger(BackendLogger.GUNICORN)