agentx-python 0.8.17__tar.gz → 0.8.19__tar.gz

{agentx_python-0.8.17/agentx_python.egg-info → agentx_python-0.8.19}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: agentx-python
-Version: 0.8.17
+Version: 0.8.19
 Summary: Official Python SDK for AgentX (https://www.agentx.so/)
 Home-page: https://github.com/AgentX-ai/AgentX-python
 Author: Robin Wang and AgentX Team

{agentx_python-0.8.17 → agentx_python-0.8.19}/agentx/monitor/client.py

@@ -187,10 +187,17 @@ class MonitorClient:
             return self._base_url[: -len(suffix)]
         return self._base_url
 
-    def _request(self, method: str, path: str, timeout: int = 30, base: Optional[str] = None, **kwargs) -> Any:
+    def _request(
+        self, method: str, path: str, timeout: int = 30, base: Optional[str] = None, retry: bool = True, **kwargs
+    ) -> Any:
+        # retry=False for non-idempotent judge-spending POSTs (sweep, coherence, portability,
+        # tuning): a client-side timeout must not fire the same LLM-billing work a second time
+        # while the first invocation is still running server-side. Same precedent as
+        # EvaluationsClient._request / analyze_run.
         url = f"{base or self._base_url}{path}"
         last_exc: Optional[Exception] = None
-        for attempt, wait in enumerate([0.0] + _RETRY_BACKOFF):
+        schedule = [0.0] + _RETRY_BACKOFF if retry else [0.0]
+        for attempt, wait in enumerate(schedule):
             if wait:
                 time.sleep(wait)
             try:
@@ -204,7 +211,7 @@ class MonitorClient:
                 raise AgentXAuthError("Invalid or missing API key")
             if resp.status_code == 422:
                 raise AgentXValidationError(resp.text)
-            if resp.status_code in _RETRYABLE_STATUS and attempt < _MAX_RETRIES - 1:
+            if retry and resp.status_code in _RETRYABLE_STATUS and attempt < _MAX_RETRIES - 1:
                 logger.debug(
                     "Retryable status %d (attempt %d)", resp.status_code, attempt + 1
                 )
@@ -421,7 +428,7 @@ class MonitorClient:
         button. Raises AgentXMonitorError if the engine has no judge key configured."""
         data = self._request(
             "POST", f"/agent-monitoring/sessions/{session_id}/coherence-check",
-            base=self._api_root(), timeout=180,
+            base=self._api_root(), timeout=180, retry=False,
         )
         return data.get("score", data) if isinstance(data, dict) else data
 
@@ -444,7 +451,7 @@ class MonitorClient:
         engines run this automatically every minute; the manual trigger exists for demos,
         tests, and backfills. Returns ``{"judged": n}``."""
         return self._request(
-            "POST", "/agent-monitoring/session-sweep/run", base=self._api_root(), timeout=300
+            "POST", "/agent-monitoring/session-sweep/run", base=self._api_root(), timeout=300, retry=False
         )
 
     # ------------------------------------------------------------------
@@ -457,7 +464,7 @@ class MonitorClient:
         plus judging, so expect tens of seconds."""
         return self._request(
             "POST", f"/agent-monitoring/traces/{trace_id}/portability",
-            base=self._api_root(), json={"modelIds": model_ids}, timeout=300,
+            base=self._api_root(), json={"modelIds": model_ids}, timeout=300, retry=False,
         )
 
     # ------------------------------------------------------------------

{agentx_python-0.8.17 → agentx_python-0.8.19}/agentx/monitor/patterns.py

@@ -114,6 +114,10 @@ class MonitorPatternClient:
             agent_ids=agent_ids,
         )
 
+    def delete(self, pattern_id: str) -> None:
+        """Delete a pattern. Its historical signals remain as history."""
+        self._client._request("DELETE", f"/agent-monitoring/patterns/{pattern_id}", base=self._client._api_root())
+
     def get(self, pattern_id: str) -> MonitorPattern:
         return self._client.get_pattern(pattern_id)
 

{agentx_python-0.8.17 → agentx_python-0.8.19}/agentx/monitor/scorer_groups.py

@@ -67,8 +67,8 @@ class ScorerGroupsClient:
         """``members``: [{"kind": "judge"|"pattern"|"custom", "refId": ..., "weight": 1, "gate": False}].
         ``online``: {"enabled": True, "sampleRate": 0.1, "alertThreshold": 5, "severity": "medium"}.
         Add ``"scope": "session", "idleSeconds": 120`` to score whole multi-turn sessions once
-        idle, instead of each sampled trace.
-        or None for offline-only."""
+        idle, instead of each sampled trace. Pass ``online=None`` (the default) for a group
+        that only grades offline dataset runs."""
         payload: Dict[str, Any] = {"name": name, "members": members}
         if description is not None:
             payload["description"] = description

{agentx_python-0.8.17 → agentx_python-0.8.19}/agentx/monitor/sessions.py

@@ -22,8 +22,9 @@ class MonitorSessionClient:
 
     def scores(self, session_id: str) -> List[dict]:
         """Session-level verdicts, newest first. ``kind`` says who scored: a session-scoped
-        online evaluator (``online-eval:<id>``) or a session-scoped scorer group
-        (``scorer-group:<id>``)."""
+        online evaluator (``online-eval:<id>``), a session-scoped scorer group
+        (``scorer-group:<id>``), or legacy ``"coherence"`` rows written before the Session
+        Baseline Judge existed - branch defensively on unknown kinds."""
         return self._client.list_session_scores(session_id)
 
     def run_sweep(self) -> dict:

{agentx_python-0.8.17 → agentx_python-0.8.19}/agentx/tracing/tracer.py

@@ -581,6 +581,14 @@ class _RetrievalRecorder:
         self.output: Any = None
 
 
+class _MemoryOpRecorder:
+    """Handle yielded by ``Tracer.trace_memory()`` - set ``output`` (what was recalled or
+    stored) inside the block."""
+
+    def __init__(self) -> None:
+        self.output: Any = None
+
+
 class _ToolCallRecorder:
     """Handle yielded by ``Tracer.trace_tool_call()`` - set ``output`` inside the block.
     ``success``/``error`` may be set manually; an exception escaping the block sets them
@@ -832,6 +840,66 @@ class Tracer:
             span_kind="retrieval",
         )
 
+    def record_memory(
+        self,
+        name: str = "Memory",
+        *,
+        operation: Optional[str] = None,
+        query: Optional[str] = None,
+        output: Any = None,
+        duration_ms: Optional[float] = None,
+        start_time: Optional[float] = None,
+        end_time: Optional[float] = None,
+    ) -> None:
+        """
+        Manually record a long-term-memory operation (a Mem0/Zep/Letta-style recall or store)
+        as a ``span_kind="memory"`` child span of the active span. ``operation`` is free text -
+        conventionally ``"read"`` or ``"write"`` - carried in the span's metadata; the kind
+        itself stays one value so dashboards and scorers can select all memory activity at once.
+        Deliberately NOT a retrieval: retrieval spans feed the RAG judges' ``{context}``
+        (knowledge grounding), while memory is recalled state - see the engine's spanKind.ts.
+        """
+        active_span = self.current_span
+        if active_span is None:
+            return
+        active_span.child_span(
+            name,
+            start_time=start_time,
+            end_time=end_time,
+            duration_ms=duration_ms,
+            input=query,
+            output=output,
+            metadata={"kind": "memory", **({"operation": operation} if operation else {})},
+            span_kind="memory",
+        )
+
+    @contextmanager
+    def trace_memory(
+        self, name: str = "Memory", *, operation: Optional[str] = None, query: Optional[str] = None
+    ) -> Iterator["_MemoryOpRecorder"]:
+        """
+        Context manager that times a memory operation and records it via
+        :meth:`record_memory` on exit::
+
+            with tracer.trace_memory("user prefs", operation="read", query=user_id) as m:
+                m.output = memory.search(user_id, question)
+        """
+        start_t = time.time()
+        recorder = _MemoryOpRecorder()
+        try:
+            yield recorder
+        finally:
+            end_t = time.time()
+            self.record_memory(
+                name,
+                operation=operation,
+                query=query,
+                output=recorder.output,
+                duration_ms=(end_t - start_t) * 1000,
+                start_time=start_t,
+                end_time=end_t,
+            )
+
     @contextmanager
     def trace_retrieval(self, name: str = "Retrieval", *, query: Optional[str] = None) -> Iterator["_RetrievalRecorder"]:
         """

{agentx_python-0.8.17 → agentx_python-0.8.19}/agentx/version.py

@@ -1,7 +1,7 @@
-VERSION = "0.8.17"
+VERSION = "0.8.19"
 
 # The AgentX-trace-eval release this SDK version is tested against - what `agentx-trace-eval`
 # installs and converges to (see agentx/cli.py). Bump together with VERSION when releasing, so
 # every published SDK names a known-good engine+dashboard pair. Users can override with
 # AGENTX_TRACE_EVAL_VERSION=<tag|latest>.
-ENGINE_VERSION = "v0.3.16"
+ENGINE_VERSION = "v0.3.19"

{agentx_python-0.8.17 → agentx_python-0.8.19/agentx_python.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: agentx-python
-Version: 0.8.17
+Version: 0.8.19
 Summary: Official Python SDK for AgentX (https://www.agentx.so/)
 Home-page: https://github.com/AgentX-ai/AgentX-python
 Author: Robin Wang and AgentX Team

{agentx_python-0.8.17 → agentx_python-0.8.19}/tests/test_selfhost_analysis_fallback.py

@@ -217,6 +217,12 @@ def test_the_fallback_request_gets_the_long_analysis_timeout():
     assert kwargs["timeout"] > 60, "a synchronous judge pass needs more than the 30s default"
     assert kwargs["json"]["judges"] == [{"model": "gpt-5.6-luna"}]
 
+    # judges=None must OMIT the key - the engine then scores with its platform default model;
+    # injecting a hosted-only default (the old "gpt-5.5") produced uncallable judges.
+    client.analyze_run(RUN)
+    _, _, kwargs = session.calls[-1]
+    assert "judges" not in kwargs["json"]
+
 
 # ---------------------------------------------------------------------------
 # Only a 404 means "wrong engine"

{agentx_python-0.8.17 → agentx_python-0.8.19}/tests/test_span_tree.py

@@ -594,6 +594,22 @@ def test_trace_tool_call_emits_real_child_span():
     assert root["tool_calls"][0]["output"] == "digital purchases are final"
 
 
+def test_trace_memory_emits_a_memory_kind_child_span():
+    tracer = make_tracer()
+    with tracer.trace("agent") as span:
+        with tracer.trace_memory("user prefs", operation="read", query="u-42") as m:
+            m.output = ["prefers window seats"]
+
+    wires = enqueued_wires(tracer)
+    child = wires[0]
+    assert child["name"] == "user prefs"
+    assert child["span_kind"] == "memory"
+    assert child["metadata"]["kind"] == "memory"
+    assert child["metadata"]["operation"] == "read"
+    assert child["input"] == "u-42"
+    assert child["output"] == ["prefers window seats"]
+
+
 def test_trace_retrieval_emits_real_child_span():
     tracer = make_tracer()
     with tracer.trace("agent") as span: