verbose live UI + tool-level SSE events + Groq default + regression tests

api/langfuse_client.py

@@ -0,0 +1,87 @@
+"""Langfuse SDK wrapper — single source of truth for the client and spans.
+
+Used by every layer (tools, analyses, plan, runtime, evals), which is why it
+lives at the api/ root rather than under any one layer.
+
+Gracefully returns no-op spans when Langfuse keys aren't configured, so local
+dev without Langfuse still works.
+"""
+
+from __future__ import annotations
+
+import logging
+from contextlib import contextmanager
+from typing import Any, Iterator
+
+from api.config import get_settings
+
+logger = logging.getLogger("nvi.langfuse")
+
+_client: Any = None
+
+
+def get_client() -> Any | None:
+    global _client
+    if _client is not None:
+        return _client
+    s = get_settings()
+    if not s.langfuse_public_key or not s.langfuse_secret_key:
+        return None
+    try:
+        from langfuse import Langfuse
+        _client = Langfuse(
+            public_key=s.langfuse_public_key,
+            secret_key=s.langfuse_secret_key,
+            host=s.langfuse_host,
+        )
+        return _client
+    except Exception as e:
+        logger.warning("langfuse init failed: %s", e)
+        return None
+
+
+class _NullSpan:
+    """No-op stand-in when Langfuse isn't configured."""
+    def update(self, **_: Any) -> None: ...
+    def end(self, **_: Any) -> None: ...
+    def __enter__(self) -> "_NullSpan": return self
+    def __exit__(self, *_: Any) -> None: ...
+
+
+@contextmanager
+def span(
+    name: str,
+    *,
+    as_type: str = "span",
+    input: Any = None,
+    metadata: dict[str, Any] | None = None,
+) -> Iterator[Any]:
+    """Open a Langfuse observation; yields the span object or a no-op.
+
+    Exceptions raised by the *body* of the `with` block propagate up
+    untouched. Only Langfuse's own setup failures fall through to a no-op
+    span — we never want to mask a real error.
+    """
+    lf = get_client()
+    if lf is None:
+        yield _NullSpan()
+        return
+    try:
+        cm = lf.start_as_current_observation(
+            name=name, as_type=as_type, input=input, metadata=metadata or {}
+        )
+    except Exception as e:
+        logger.warning("langfuse setup for %s failed: %s", name, e)
+        yield _NullSpan()
+        return
+    with cm as s:
+        yield s
+
+
+def flush() -> None:
+    lf = get_client()
+    if lf is not None:
+        try:
+            lf.flush()
+        except Exception:
+            pass