agentx-python 0.8.16__tar.gz → 0.8.18__tar.gz

{agentx_python-0.8.16/agentx_python.egg-info → agentx_python-0.8.18}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: agentx-python
-Version: 0.8.16
+Version: 0.8.18
 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-0.8.16 → agentx_python-0.8.18}/agentx/evaluations/datasets.py

@@ -48,7 +48,7 @@ class DatasetBuilder:
             "questions": [],
         }
         # LLM-as-judge overrides for this dataset's own grading config. Omit either to keep the
-        # server default (raw prompt template / OpenAI gpt-5.5, see EVALUATIONS.md). judge_model
+        # server default (raw prompt template / gpt-5.6-luna, see EVALUATIONS.md). judge_model
         # must be one of client.evaluations.list_models() (OpenAI or Anthropic).
         if judge_prompt is not None:
             self._payload["judgePrompt"] = judge_prompt

{agentx_python-0.8.16 → agentx_python-0.8.18}/agentx/evaluations/evaluation_settings.py

@@ -45,7 +45,7 @@ class EvaluationSettingsBuilder:
             "evaluationCriteria": evaluation_criteria,
         }
         # LLM-as-judge overrides. Omit either to keep the server default (raw prompt template /
-        # OpenAI gpt-5.5, see EVALUATIONS.md). judge_model must be one of
+        # gpt-5.6-luna, see EVALUATIONS.md). judge_model must be one of
         # client.evaluations.list_models() (OpenAI or Anthropic).
         if judge_prompt is not None:
             self._payload["judgePrompt"] = judge_prompt

{agentx_python-0.8.16 → agentx_python-0.8.18}/agentx/evaluations/models.py

@@ -113,8 +113,8 @@ class EvaluationSettings(BaseModel):
     evaluation_criteria: Optional[str] = Field(default=None, alias="evaluationCriteria")
     # Custom code scorers on this scorer's offline profile - [{ id, name, code, enabled }].
     code_scorers: Optional[List[Dict[str, Any]]] = Field(default=None, alias="codeScorers")
-    # LLM-as-judge overrides. None means "use the server default" (raw prompt template / OpenAI
-    # gpt-5.5). See client.evaluations.settings.builder(judge_prompt=..., judge_model=...).
+    # LLM-as-judge overrides. None means "use the server default" (raw prompt template /
+    # gpt-5.6-luna). See client.evaluations.settings.builder(judge_prompt=..., judge_model=...).
     judge_prompt: Optional[str] = Field(default=None, alias="judgePrompt")
     judge_model: Optional[str] = Field(default=None, alias="judgeModel")
     status: str = "published"

{agentx_python-0.8.16 → agentx_python-0.8.18}/agentx/evaluations/runner.py

@@ -63,9 +63,6 @@ _ANALYSIS_LEVEL_LABELS = {
     "l4_final_reduce": "writing final report",
 }
 
-_DEFAULT_JUDGE_MODEL = "gpt-5.5"
-
-
 class GateResult:
     """Wire result of the CI gate (GET /runs/:id/gate) with attribute access for the fields a
     CI script actually branches on."""
@@ -456,16 +453,16 @@ class EvaluationRunContext:
         Args:
             mode: "auto" (default), "sync", or "batch" - how item scoring executes server-side.
             quality_mode: "quality_first" or "balanced" - how many items get a second/third judge.
-            judges: 1-3 model ids, e.g. ``["gpt-5.5", "claude-opus-4-8"]``. Defaults to a single
-                judge, ``["gpt-5.5"]``, rather than the dashboard's 3-judge default - SDK runs are
-                typically lighter-weight, quick-start evaluations.
+            judges: 1-3 model ids from ``client.evaluations.list_models()``, e.g.
+                ``["gpt-5.6-luna", "claude-opus-4-8"]``. Omit to let the engine score with its
+                platform default model (a single judge, rather than the dashboard's 3-judge
+                default - SDK runs are typically lighter-weight, quick-start evaluations).
             poll_interval: seconds between status checks while waiting.
             timeout: give up waiting after this many seconds (the job keeps running server-side;
                 call ``get_report()`` later to check on it).
         """
         if judges is not None and not (1 <= len(judges) <= 3):
             raise ValueError("judges must contain 1-3 model ids")
-        resolved_judges = judges if judges is not None else [_DEFAULT_JUDGE_MODEL]
 
         _say()
         with Spinner("Analyzing - AI is reviewing your results") as spinner:
@@ -474,7 +471,7 @@ class EvaluationRunContext:
                     self._run.run_id,
                     mode=mode,
                     quality_mode=quality_mode,
-                    judges=resolved_judges,
+                    judges=judges,
                 )
                 deadline = time.monotonic() + timeout
                 status = self._client.get_analysis_status(self._run.run_id)

{agentx_python-0.8.16 → agentx_python-0.8.18}/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
 
@@ -429,6 +436,24 @@ class MonitorClient:
         data = self._request("GET", f"/ingest/sessions/{session_id}/spans", base=self._api_root())
         return data.get("spans", []) if isinstance(data, dict) else data
 
+    def list_session_scores(self, session_id: str) -> List[dict]:
+        """Every session-level verdict on the session, newest first: session-scoped online
+        evaluators (kind ``online-eval:<id>``), session-scoped scorer groups
+        (``scorer-group:<id>``), and legacy coherence rows."""
+        data = self._request(
+            "GET", f"/agent-monitoring/sessions/{session_id}/scores", base=self._api_root()
+        )
+        return data.get("scores", []) if isinstance(data, dict) else data
+
+    def run_session_sweep(self) -> dict:
+        """Run the idle-session sweep once, now - the tick that scores quiet multi-turn
+        sessions with every enabled session-scoped evaluator and scorer group. Production
+        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, retry=False
+        )
+
     # ------------------------------------------------------------------
     # Model portability (self-host): replay a trace's input against other models
     # ------------------------------------------------------------------
@@ -439,7 +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-0.8.16 → agentx_python-0.8.18}/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-0.8.16 → agentx_python-0.8.18}/agentx/monitor/scorer_groups.py

@@ -65,8 +65,10 @@ class ScorerGroupsClient:
         online: Optional[Dict[str, Any]] = None,
     ) -> ScorerGroup:
         """``members``: [{"kind": "judge"|"pattern"|"custom", "refId": ..., "weight": 1, "gate": False}].
-        ``online``: {"enabled": True, "sampleRate": 0.1, "alertThreshold": 5, "severity": "medium"}
-        or None for offline-only."""
+        ``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. 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-0.8.16 → agentx_python-0.8.18}/agentx/monitor/sessions.py

@@ -19,3 +19,16 @@ class MonitorSessionClient:
     def spans(self, session_id: str) -> List[dict]:
         """Every span in the session (roots and children), oldest first."""
         return self._client.list_session_spans(session_id)
+
+    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>``), 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:
+        """Trigger the idle-session sweep once (normally automatic, every minute) - scores
+        idle multi-turn sessions with every enabled session-scoped evaluator and scorer
+        group. Returns ``{"judged": n}``."""
+        return self._client.run_session_sweep()

{agentx_python-0.8.16 → agentx_python-0.8.18}/agentx/version.py

@@ -1,7 +1,7 @@
-VERSION = "0.8.16"
+VERSION = "0.8.18"
 
 # 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 = "v0.3.13"
+ENGINE_VERSION = "v0.3.18"

{agentx_python-0.8.16 → agentx_python-0.8.18/agentx_python.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: agentx-python
-Version: 0.8.16
+Version: 0.8.18
 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-0.8.16 → agentx_python-0.8.18}/tests/test_integrations.py

@@ -599,7 +599,9 @@ def test_crewai_captures_real_per_task_timing_via_event_bus():
     durations are provably unequal — the old "divide latency evenly across
     tasks" approximation would have reported them as identical.
     """
-    crewai = pytest.importorskip("crewai")
+    # exc_type=Exception: on older Pythons crewai can raise TypeError (PEP 604 syntax) at
+    # import time, and a broken optional integration should skip this test, not fail it.
+    crewai = pytest.importorskip("crewai", exc_type=Exception)
     from crewai.events.event_bus import crewai_event_bus
     from crewai.events.types.task_events import TaskCompletedEvent, TaskStartedEvent
     from crewai.tasks.task_output import TaskOutput

{agentx_python-0.8.16 → agentx_python-0.8.18}/tests/test_selfhost_analysis_fallback.py

@@ -210,12 +210,18 @@ def test_the_fallback_request_gets_the_long_analysis_timeout():
         {("POST", f"{API_ROOT}/evaluate/analyze/{RUN}"): FakeResponse(200, {"status": "completed"})}
     )
 
-    client.analyze_run(RUN, judges=["gpt-5.5"])
+    client.analyze_run(RUN, judges=["gpt-5.6-luna"])
 
     method, url, kwargs = session.calls[-1]
     assert url == f"{API_ROOT}/evaluate/analyze/{RUN}"
     assert kwargs["timeout"] > 60, "a synchronous judge pass needs more than the 30s default"
-    assert kwargs["json"]["judges"] == [{"model": "gpt-5.5"}]
+    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"]
 
 
 # ---------------------------------------------------------------------------