diff --git a/README.md b/README.md
index 115073cd0..7fed05bba 100644
--- a/README.md
+++ b/README.md
@@ -516,6 +516,10 @@ Run `curl -fsSL https://raw.githubusercontent.com/jaylfc/taOS/master/scripts/ins
 | OS packages added | python3 + venv + pip, git, curl, ca-certificates, libtorrent-rasterbar (model torrent mesh), Node.js 22 (qmd + SPA build), sqlite3, libsqlcipher (encrypted secrets), vulkan-tools (hardware detection), postgresql (LiteLLM virtual keys) |
 | User accounts created | The distro `postgres` system user is created when PostgreSQL is installed. A `litellm` Postgres role and database are created for virtual-key management. Everything else runs as the user who ran the installer. |
 
+### Hailo-10H install (`scripts/install-hailo.sh`)
+
+Installs `hailo-ollama` (the Hailo-10H LLM runtime) at a pinned ref, listening on port **7836** (port 8000 is the Django slot in taOS port hygiene and is already probed as a llama-cpp/vllm candidate). Before installing, the script checks for a pre-existing upstream hailo-ollama (something answering `GET /api/tags` on `0.0.0.0:8000`, Hailo's own default). If one is found it reports plainly what was detected and exits, leaving the existing instance untouched rather than silently building a second server that would compete for the accelerator and go unseen by taOS.
+
 ### Worker install (`scripts/install-worker.sh`)
 
 Linux uses a two-phase install. Phase 1 runs on the bare host; phase 2 runs inside the `taos-worker` LXC.
diff --git a/changelog.d/2083-hailo-preinstall-detect.md b/changelog.d/2083-hailo-preinstall-detect.md
new file mode 100644
index 000000000..a86ca174e
--- /dev/null
+++ b/changelog.d/2083-hailo-preinstall-detect.md
@@ -0,0 +1,10 @@
+### Changed
+
+- The Hailo installer (`scripts/install-hailo.sh`) now checks for a pre-existing
+  hailo-ollama instance on upstream port 8000 before any install or systemd
+  mutation. When found (a server answering `GET /api/tags` with `"models"`), the
+  script prints what was detected, reports that it would have built a second
+  server on port 7836, and exits — leaving the existing instance untouched. This
+  prevents the silent coexistence bug from issue #2083 where an upstream
+  hailo-ollama (running on its default `0.0.0.0:8000`) would go unseen while
+  taOS built a second server and installed no systemd unit.
diff --git a/scripts/install-hailo.sh b/scripts/install-hailo.sh
index 8b6f39701..0739bcf7c 100755
--- a/scripts/install-hailo.sh
+++ b/scripts/install-hailo.sh
@@ -240,6 +240,28 @@ detect_hailo() {
     return 0
 }
 
+# -------- (1.5) pre-install detection --------------------------------------
+
+# Checks for a pre-existing hailo-ollama instance running on upstream port 8000
+# (Hailo's own default / the Ollama-compatible tags endpoint). When found, logs
+# what was detected and exits — taOS must not silently build a second server.
+# Port 8000 is the Django slot in taOS port hygiene and is already probed as a
+# llama-cpp/vllm candidate, so a coexisting server there also makes those probes
+# ambiguous. See issue #2083.
+detect_preexisting_hailoollama() {
+    local url="http://0.0.0.0:8000/api/tags"
+    local tags
+    tags="$(curl -fs "$url" 2>/dev/null || true)"
+    if [[ -n "$tags" ]] && grep -q '"models"' <<<"$tags"; then
+        warn "pre-existing hailo-ollama detected on :8000 (Hailo's default upstream)"
+        log "response from $url:"
+        echo "$tags" | sed 's/^/    /' || true
+        warn "This installer would have built a second server on port $HAILO_OLLAMA_PORT."
+        log "The existing instance on :8000 will be left alone (not modified by this script)."
+        exit 0
+    fi
+}
+
 # -------- (2) HailoRT + firmware -----------------------------------------
 
 ensure_hailort() {
@@ -471,6 +493,9 @@ main() {
             ;;
     esac
 
+    # Pre-install: check for a pre-existing hailo-ollama on upstream port 8000.
+    detect_preexisting_hailoollama
+
     resolve_target
 
     if already_installed; then