@melaya/runner 1.0.44 → 1.0.46

package/dist/localRagIngest.py

@@ -136,12 +136,57 @@ async def _ingest(args) -> int:
             # LM Studio exposes an OpenAI-compatible /v1; reuse the OpenAI
             # client pointed at its local endpoint. Dim default = 768 (the
             # FE picker should report the real dim per loaded model).
+            # base_url MUST be passed at construction (the OpenAI 1.x
+            # AsyncClient bakes the URL into its httpx client during
+            # __init__; setting `client.base_url = ...` after the fact is a
+            # no-op and the next request resolves an empty hostname →
+            # `[Errno 8] nodename nor servname provided`).
             embedder = OpenAITextEmbedding(
                 api_key="lmstudio",
                 model_name=args.embedder_model,
                 dimensions=768,
+                base_url="http://localhost:1234/v1",
             )
-            embedder.client.base_url = "http://localhost:1234/v1"
+            # Diagnostic: surface the URLs the httpx client will actually
+            # hit. Prior runs were failing with `[Errno 8] nodename nor
+            # servname provided` — the high-level `.base_url` attribute
+            # was being set but the underlying httpx client kept the
+            # default `api.openai.com` from construction. With the kwarg
+            # path these should both be `localhost:1234/v1`.
+            try:
+                _cb = getattr(embedder.client, "base_url", "?")
+                _hb = getattr(getattr(embedder.client, "_client", None), "base_url", "?")
+                _emit("lmstudio_init", client_base_url=str(_cb), httpx_base_url=str(_hb))
+            except Exception as _exc:
+                _emit("lmstudio_init", error=f"{type(_exc).__name__}: {_exc}")
+            # Probe the LM Studio HTTP server BEFORE the first embed call
+            # so a misconfigured/off server surfaces as one clear error
+            # rather than 240 cryptic per-file failures.
+            try:
+                import urllib.request, json as _json
+                with urllib.request.urlopen(
+                    "http://localhost:1234/v1/models", timeout=3
+                ) as _r:
+                    _models = _json.loads(_r.read().decode("utf-8"))
+                    _names = [m.get("id") for m in _models.get("data", [])]
+                    _emit("lmstudio_probe", ok=True, models=_names)
+                    if args.embedder_model not in _names:
+                        _emit("warning", reason=(
+                            f"LM Studio is reachable but model "
+                            f"'{args.embedder_model}' is not loaded. "
+                            f"Loaded: {_names}"
+                        ))
+            except Exception as _exc:
+                _emit("lmstudio_probe", ok=False,
+                      error=f"{type(_exc).__name__}: {_exc}")
+                _emit("error", reason=(
+                    f"Cannot reach LM Studio at http://localhost:1234. "
+                    f"Make sure the LM Studio app's local server is started "
+                    f"(Developer → Start Server) and the model "
+                    f"'{args.embedder_model}' is loaded. Underlying error: "
+                    f"{type(_exc).__name__}: {_exc}"
+                ))
+                return 6
             # LM Studio's /v1/embeddings rejects the OpenAI-specific
             # `dimensions` kwarg (only OpenAI text-embedding-3 supports
             # output-dim reduction). Strip it before each call so the

package/localRagIngest.py

@@ -136,12 +136,57 @@ async def _ingest(args) -> int:
             # LM Studio exposes an OpenAI-compatible /v1; reuse the OpenAI
             # client pointed at its local endpoint. Dim default = 768 (the
             # FE picker should report the real dim per loaded model).
+            # base_url MUST be passed at construction (the OpenAI 1.x
+            # AsyncClient bakes the URL into its httpx client during
+            # __init__; setting `client.base_url = ...` after the fact is a
+            # no-op and the next request resolves an empty hostname →
+            # `[Errno 8] nodename nor servname provided`).
             embedder = OpenAITextEmbedding(
                 api_key="lmstudio",
                 model_name=args.embedder_model,
                 dimensions=768,
+                base_url="http://localhost:1234/v1",
             )
-            embedder.client.base_url = "http://localhost:1234/v1"
+            # Diagnostic: surface the URLs the httpx client will actually
+            # hit. Prior runs were failing with `[Errno 8] nodename nor
+            # servname provided` — the high-level `.base_url` attribute
+            # was being set but the underlying httpx client kept the
+            # default `api.openai.com` from construction. With the kwarg
+            # path these should both be `localhost:1234/v1`.
+            try:
+                _cb = getattr(embedder.client, "base_url", "?")
+                _hb = getattr(getattr(embedder.client, "_client", None), "base_url", "?")
+                _emit("lmstudio_init", client_base_url=str(_cb), httpx_base_url=str(_hb))
+            except Exception as _exc:
+                _emit("lmstudio_init", error=f"{type(_exc).__name__}: {_exc}")
+            # Probe the LM Studio HTTP server BEFORE the first embed call
+            # so a misconfigured/off server surfaces as one clear error
+            # rather than 240 cryptic per-file failures.
+            try:
+                import urllib.request, json as _json
+                with urllib.request.urlopen(
+                    "http://localhost:1234/v1/models", timeout=3
+                ) as _r:
+                    _models = _json.loads(_r.read().decode("utf-8"))
+                    _names = [m.get("id") for m in _models.get("data", [])]
+                    _emit("lmstudio_probe", ok=True, models=_names)
+                    if args.embedder_model not in _names:
+                        _emit("warning", reason=(
+                            f"LM Studio is reachable but model "
+                            f"'{args.embedder_model}' is not loaded. "
+                            f"Loaded: {_names}"
+                        ))
+            except Exception as _exc:
+                _emit("lmstudio_probe", ok=False,
+                      error=f"{type(_exc).__name__}: {_exc}")
+                _emit("error", reason=(
+                    f"Cannot reach LM Studio at http://localhost:1234. "
+                    f"Make sure the LM Studio app's local server is started "
+                    f"(Developer → Start Server) and the model "
+                    f"'{args.embedder_model}' is loaded. Underlying error: "
+                    f"{type(_exc).__name__}: {_exc}"
+                ))
+                return 6
             # LM Studio's /v1/embeddings rejects the OpenAI-specific
             # `dimensions` kwarg (only OpenAI text-embedding-3 supports
             # output-dim reduction). Strip it before each call so the

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@melaya/runner",
-  "version": "1.0.44",
+  "version": "1.0.46",
   "description": "Run Melaya AI pipelines locally with your own LM Studio or Ollama models",
   "license": "UNLICENSED",
   "private": false,