@pentatonic-ai/ai-agent-sdk 0.8.1 → 0.8.2

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@pentatonic-ai/ai-agent-sdk",
-  "version": "0.8.1",
+  "version": "0.8.2",
   "description": "TES SDK — LLM observability and lifecycle tracking via Pentatonic Thing Event System. Track token usage, tool calls, and conversations. Manage things through event-sourced lifecycle stages with AI enrichment and vector search.",
   "type": "module",
   "main": "./dist/index.cjs",

package/packages/memory-engine/compat/server.py

@@ -416,8 +416,14 @@ async def health():
         "engine": "pentatonic-memory-engine",
         "layers": {},
     }
-    # NV-Embed exposes /health alongside /v1/embeddings.
-    nv_embed_health = NV_EMBED_URL.replace("/v1/embeddings", "/health")
+    # NV-Embed (or the upstream gateway) exposes /health at the host root.
+    # Use urlparse so we swap *just* the path component instead of doing a
+    # string replace — that breaks the moment NV_EMBED_URL is /v1/embed,
+    # /v1/embeddings, or bare-host. The probe is informational only; gateways
+    # that return non-200 on root-/health don't block engine operation.
+    from urllib.parse import urlparse, urlunparse
+    _u = urlparse(NV_EMBED_URL)
+    nv_embed_health = urlunparse((_u.scheme, _u.netloc, "/health", "", "", ""))
 
     import asyncio
     l2_v, l4_v, l5_v, l6_v, nv_v, l3_v = await asyncio.gather(

package/packages/memory-engine/tests/test_compat_nv_embed_probe.py

@@ -0,0 +1,48 @@
+"""Regression test for the nv_embed /health probe URL construction.
+
+Before v0.8.1 the probe was built via str.replace("/v1/embeddings", "/health"),
+which silently produced the wrong URL when NV_EMBED_URL was anything other
+than ".../v1/embeddings" (e.g. /v1/embed for the Pentatonic AI Gateway, or
+a bare host). The probe would end up GET-ing the embed endpoint itself and
+the gateway would return 401 — making /health falsely report a degraded
+nv_embed layer even when embeddings worked fine.
+
+The fix replaces the substring replace with a urlparse-based path swap.
+"""
+
+from urllib.parse import urlparse, urlunparse
+
+
+def probe_url(nv_embed_url: str) -> str:
+    """Re-implementation of the probe URL builder from compat/server.py.
+    Kept tiny + isolated so it can be unit-tested without booting FastAPI."""
+    u = urlparse(nv_embed_url)
+    return urlunparse((u.scheme, u.netloc, "/health", "", "", ""))
+
+
+def test_probe_with_v1_embeddings_path():
+    assert probe_url("https://nv-embed:8041/v1/embeddings") == "https://nv-embed:8041/health"
+
+
+def test_probe_with_v1_embed_path():
+    """The bug case — gateway moved to /v1/embed and old replace() left URL untouched."""
+    assert probe_url("https://lambda-gateway.pentatonic.com/v1/embed") == (
+        "https://lambda-gateway.pentatonic.com/health"
+    )
+
+
+def test_probe_with_bare_host():
+    """No path at all — provider supplies path_default, but probe just wants /health."""
+    assert probe_url("https://lambda-gateway.pentatonic.com") == (
+        "https://lambda-gateway.pentatonic.com/health"
+    )
+
+
+def test_probe_strips_query_and_fragment():
+    assert probe_url("https://gw.example.com/v1/embeddings?token=x#frag") == (
+        "https://gw.example.com/health"
+    )
+
+
+def test_probe_preserves_port():
+    assert probe_url("http://nv-embed:8041/v1/embeddings") == "http://nv-embed:8041/health"