agentx-python 0.6.34__tar.gz → 0.6.36__tar.gz

{agentx_python-0.6.34 → agentx_python-0.6.36}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: agentx-python
-Version: 0.6.34
+Version: 0.6.36
 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-0.6.34 → agentx_python-0.6.36}/agentx/agentx.py

@@ -20,10 +20,12 @@ class AgentX:
         if self.api_key and not os.getenv("AGENTX_API_KEY"):
             os.environ["AGENTX_API_KEY"] = self.api_key
 
-        # base_url overrides AGENTX_API_BASE_URL env var (and the SDK default)
+        # base_url overrides AGENTX_API_BASE_URL env var (and the SDK default). It is
+        # deliberately NOT written back into os.environ: the constructor used to do that, which
+        # made the last-constructed client silently re-point every other client in the process
+        # (deep-dive round 3, bug #1). Each sub-client below receives this value explicitly and
+        # captures it at construction instead.
         self.base_url = base_url or os.getenv("AGENTX_API_BASE_URL")
-        if self.base_url:
-            os.environ["AGENTX_API_BASE_URL"] = self.base_url
 
         self.workspace_id = workspace_id or os.getenv("AGENTX_WORKSPACE_ID")
 
@@ -58,29 +60,29 @@ class AgentX:
 
         # Report real, after-the-fact outcomes ("the ticket got reopened") against traces - the
         # ground truth behind the dashboard's Judge Calibration card. Self-host only.
-        self.outcomes = OutcomesClient(api_key=self.api_key)
+        self.outcomes = OutcomesClient(api_key=self.api_key, base_url=self.base_url)
 
         from agentx.projects import ProjectsClient
 
         # Project CRUD (self-host): isolated tenants with their own API keys (P1.1).
-        self.projects = ProjectsClient(api_key=self.api_key)
+        self.projects = ProjectsClient(api_key=self.api_key, base_url=self.base_url)
 
         from agentx.traces import TracesClient
 
         # The read side of tracing: trace-by-id detail and paginated listing (P1.2).
-        self.traces = TracesClient(api_key=self.api_key)
+        self.traces = TracesClient(api_key=self.api_key, base_url=self.base_url)
 
         from agentx.export import ExportClient
 
         # Bulk NDJSON egress for backup/migration (P2.1): manifest, per-entity streaming, and
         # directory dumps. Self-host only.
-        self.export = ExportClient(api_key=self.api_key)
+        self.export = ExportClient(api_key=self.api_key, base_url=self.base_url)
 
         from agentx.feedback import FeedbackClient
 
         # Forward end-user votes ("up"/"down") on traced responses - a "down" raises a signal
         # directly, and every vote feeds Judge Calibration alongside outcomes. Self-host only.
-        self.feedback = FeedbackClient(api_key=self.api_key)
+        self.feedback = FeedbackClient(api_key=self.api_key, base_url=self.base_url)
 
         _ingest_client = IngestClient(
             api_key=self.api_key,
@@ -92,13 +94,23 @@ class AgentX:
 
     @classmethod
     def from_env(cls) -> "AgentX":
-        """Create an AgentX client using AGENTX_API_KEY (and optionally AGENTX_API_BASE_URL) from the environment."""
-        return cls()
+        """Create an AgentX client from the environment: AGENTX_API_KEY plus, for the base URL,
+        the first of AGENTX_API_BASE_URL / AGENTX_SELFHOST_BASE_URL / BASE_URL that is set.
+        The fallbacks match the conventions the self-host samples and .env files already use,
+        so from_env works wherever an explicit AgentX(base_url=...) would."""
+        import os
+
+        base_url = (
+            os.getenv("AGENTX_API_BASE_URL")
+            or os.getenv("AGENTX_SELFHOST_BASE_URL")
+            or os.getenv("BASE_URL")
+        )
+        return cls(base_url=base_url) if base_url else cls()
 
     def get_agent(self, id: str) -> Agent:
-        url = f"{api_base()}/access/agents/{id}"
+        url = f"{self.base_url or api_base()}/access/agents/{id}"
         # Make a GET request to the AgentX API
-        response = requests.get(url, headers=get_headers())
+        response = requests.get(url, headers=get_headers(self.api_key))
         # Check if response was successful
         if response.status_code == 200:
             return Agent(**response.json())
@@ -106,9 +118,9 @@ class AgentX:
             raise Exception(f"Failed to retrieve agent: {response.reason}")
 
     def list_agents(self) -> List[Agent]:
-        url = f"{api_base()}/access/agents"
+        url = f"{self.base_url or api_base()}/access/agents"
         # Make a GET request to the AgentX API
-        response = requests.get(url, headers=get_headers())
+        response = requests.get(url, headers=get_headers(self.api_key))
         # Check if response was successful
         if response.status_code == 200:
             return [Agent(**agent) for agent in response.json()]
@@ -146,7 +158,7 @@ class AgentX:
         """
         from agentx.exceptions import AgentXAPIError, AgentXAuthError, AgentXConnectionError
 
-        base = api_base()
+        base = (self.base_url or api_base()).rstrip("/")
         # /monitor/patterns: the cheapest key-authenticated endpoint that exists on both the
         # hosted API and the self-host engine's SDK-facing router.
         url = f"{base}/monitor/patterns"
@@ -173,8 +185,8 @@ class AgentX:
 
     def get_profile(self):
         """Get the current user's profile information."""
-        url = f"{api_base()}/access/getProfile"
-        response = requests.get(url, headers=get_headers())
+        url = f"{self.base_url or api_base()}/access/getProfile"
+        response = requests.get(url, headers=get_headers(self.api_key))
         if response.status_code == 200:
             return response.json()
         else:

{agentx_python-0.6.34 → agentx_python-0.6.36}/agentx/evaluations/client.py

@@ -59,6 +59,15 @@ class AgentXValidationError(AgentXEvaluationsError):
     pass
 
 
+def _resolve_scorer_id(scorer_id: Optional[str], evaluation_settings_id: Optional[str]) -> Optional[str]:
+    """One grader, two spellings: ``scorer_id`` is the post-consolidation name for what the wire
+    still calls ``evaluationSettingsId`` (the ids are identical by design). Both kwargs are
+    accepted everywhere a run picks its grader; passing both with different values is a bug."""
+    if scorer_id and evaluation_settings_id and scorer_id != evaluation_settings_id:
+        raise ValueError("Pass either scorer_id or evaluation_settings_id (they are the same id), not two different ids")
+    return scorer_id or evaluation_settings_id
+
+
 class EvaluationsClient:
     def __init__(
         self,
@@ -100,11 +109,21 @@ class EvaluationsClient:
         from agentx.evaluations.prompts import PromptClient
 
         self.datasets = DatasetClient(self)
-        self.settings = EvaluationSettingsClient(self)
+        # Legacy view of an LLM Judge Scorer's offline profile - constructed lazily so its
+        # DeprecationWarning fires on first USE, not for every client that never touches it.
+        self._settings: "EvaluationSettingsClient | None" = None
         self.prompts = PromptClient(self)
         from agentx.evaluations.tool_schemas import ToolSchemaClient
         self.tool_schemas = ToolSchemaClient(self)
 
+    @property
+    def settings(self) -> "EvaluationSettingsClient":
+        if self._settings is None:
+            from agentx.evaluations.evaluation_settings import EvaluationSettingsClient
+
+            self._settings = EvaluationSettingsClient(self)
+        return self._settings
+
     # ------------------------------------------------------------------
     # Low-level HTTP
     # ------------------------------------------------------------------
@@ -275,10 +294,16 @@ class EvaluationsClient:
         dataset_id: str,
         subject: EvaluationSubject,
         python_version: Optional[str] = None,
+        scorer_id: Optional[str] = None,
         evaluation_settings_id: Optional[str] = None,
     ) -> EvaluationRun:
+        """``scorer_id`` names the LLM Judge Scorer grading this run (its id doubles as the
+        wire's ``evaluationSettingsId``). ``evaluation_settings_id`` is the pre-consolidation
+        alias and keeps working."""
         from agentx.version import VERSION
 
+        grader_id = _resolve_scorer_id(scorer_id, evaluation_settings_id)
+
         payload = {
             "datasetId": dataset_id,
             "evaluationSubject": subject.model_dump(by_alias=True, exclude_none=True),
@@ -290,8 +315,8 @@ class EvaluationsClient:
                 "pythonVersion": python_version or _python_version(),
             },
         }
-        if evaluation_settings_id:
-            payload["evaluationSettingsId"] = evaluation_settings_id
+        if grader_id:
+            payload["evaluationSettingsId"] = grader_id
         data = self._request("POST", "/runs", json=self._with_workspace(payload))
         return EvaluationRun(**data)
 

{agentx_python-0.6.34 → agentx_python-0.6.36}/agentx/evaluations/datasets.py

@@ -290,6 +290,7 @@ class DatasetClient:
         rouge_score: bool = False,
         similarity_model: Optional[str] = None,
         sovereignty_models: Optional[List[str]] = None,
+        code_scorers: Optional[List[Dict[str, Any]]] = None,
     ) -> DatasetBuilder:
         return DatasetBuilder(
             self._client,
@@ -307,6 +308,9 @@ class DatasetClient:
             rouge_score=rouge_score,
             similarity_model=similarity_model,
             sovereignty_models=sovereignty_models,
+            # Was documented (evaluation/code-scorers.mdx) but not forwarded - DatasetBuilder
+            # itself always accepted it. Fixed with the judge-scorer unification.
+            code_scorers=code_scorers,
         )
 
     def from_csv(self, path: str, name: str, **kwargs) -> DatasetBuilder:

{agentx_python-0.6.34 → agentx_python-0.6.36}/agentx/evaluations/evaluation_settings.py

@@ -1,6 +1,7 @@
 from __future__ import annotations
 
 import logging
+import warnings
 from typing import Any, Dict, List, Optional, TYPE_CHECKING
 
 from agentx.evaluations.models import EvaluationSettings
@@ -81,10 +82,25 @@ class EvaluationSettingsBuilder:
 
 
 class EvaluationSettingsClient:
-    """Thin wrapper surfaced as client.evaluations.settings."""
+    """Thin wrapper surfaced as ``client.evaluations.settings``.
+
+    Note: an evaluation-settings record is the judge rubric + OFFLINE profile of an
+    **LLM Judge Scorer** - the unified entity at ``client.monitor.judge_scorers``, which also
+    carries the optional online (live-traffic) profile. This client keeps working unchanged;
+    prefer ``judge_scorers`` for new code so both profiles live in one place -
+    ``client.monitor.judge_scorers.builder(...)`` has the same snake_case ergonomics as the
+    builder below, plus tool_context, thresholds, and the live profile in one call."""
 
     def __init__(self, client: "EvaluationsClient"):
         self._client = client
+        # Soft deprecation: hidden by default (DeprecationWarning), visible under -W or pytest.
+        warnings.warn(
+            "client.evaluations.settings is the legacy view of an LLM Judge Scorer's offline "
+            "profile; prefer client.monitor.judge_scorers, which manages the judge rubric, "
+            "offline profile, and online (live-traffic) profile as one entity.",
+            DeprecationWarning,
+            stacklevel=3,
+        )
 
     def builder(
         self,

{agentx_python-0.6.34 → agentx_python-0.6.36}/agentx/evaluations/models.py

@@ -94,7 +94,7 @@ class Dataset(BaseModel):
 
 class EvaluationSettings(BaseModel):
     """A standalone, reusable grading config - no dataset/questions attached.
-    Created via ``client.evaluations.settings.builder(...).publish()`` and run
+    Created via ``client.monitor.judge_scorers.builder(...).publish()`` (or the legacy ``client.evaluations.settings.builder``) and run
     against any dataset by passing its id as ``evaluation_settings_id`` to
     ``client.evaluations.run(...)``."""
 

{agentx_python-0.6.34 → agentx_python-0.6.36}/agentx/evaluations/runner.py

@@ -421,10 +421,15 @@ class EvaluationsRunner:
     def __init__(self, client: EvaluationsClient):
         self._client = client
         self.datasets = client.datasets
-        self.settings = client.settings
         self.prompts = client.prompts
         self.tool_schemas = client.tool_schemas
 
+    @property
+    def settings(self):
+        # Deferred: touching client.settings eagerly here would fire its legacy-surface
+        # DeprecationWarning for every runner, used or not.
+        return self._client.settings
+
     def list_models(self, provider: Optional[str] = None) -> List[ModelInfo]:
         """List the LLM models AgentX supports - the same set selectable for
         the Sovereignty & Portability Index. Pass ``provider`` (e.g. "Google")
@@ -482,24 +487,24 @@ class EvaluationsRunner:
         self,
         dataset_id: str,
         subject: Union[Dict[str, Any], EvaluationSubject],
+        scorer_id: Optional[str] = None,
         evaluation_settings_id: Optional[str] = None,
     ) -> EvaluationRunContext:
-        """Start a run of ``dataset_id`` against ``subject``. Pass
-        ``evaluation_settings_id`` to grade against a standalone, reusable
-        config (created via ``client.evaluations.settings.builder(...)``)
-        instead of the dataset's own default config."""
+        """Start a run of ``dataset_id`` against ``subject``. Pass ``scorer_id`` (an LLM Judge
+        Scorer's id, e.g. from ``client.monitor.judge_scorers``) to grade with a specific
+        scorer instead of the dataset's default. ``evaluation_settings_id`` is the
+        pre-consolidation alias for the same id and keeps working."""
+        from agentx.evaluations.client import _resolve_scorer_id
+
         if isinstance(subject, dict):
             subject = EvaluationSubject(**subject)
 
+        grader_id = _resolve_scorer_id(scorer_id, evaluation_settings_id)
         dataset = self._client.get_dataset(dataset_id)
         evaluation_settings = (
-            self._client.get_evaluation_settings(evaluation_settings_id)
-            if evaluation_settings_id
-            else None
-        )
-        run = self._client.init_run(
-            dataset_id, subject, evaluation_settings_id=evaluation_settings_id
+            self._client.get_evaluation_settings(grader_id) if grader_id else None
         )
+        run = self._client.init_run(dataset_id, subject, scorer_id=grader_id)
         logger.info(
             "Started evaluation run %s on dataset %s (%d case(s), %d repetition(s))",
             run.run_id,

{agentx_python-0.6.34 → agentx_python-0.6.36}/agentx/export.py

@@ -36,13 +36,15 @@ class ExportClient:
     (``pg_dump`` / SQLite file copy). There is deliberately no blind row-import endpoint.
     """
 
-    def __init__(self, api_key: Optional[str] = None):
+    def __init__(self, api_key: Optional[str] = None, base_url: Optional[str] = None):
         self._api_key = api_key
+        # Captured once at construction (deep-dive round 3, bug #1).
+        self._base_url = (base_url or api_base()).rstrip("/")
 
     def manifest(self) -> List[Dict[str, Any]]:
         """The exportable entities with live row counts: ``[{entity, rows, path}, ...]``."""
         resp = requests.get(
-            f"{api_base()}/export", headers=get_headers(self._api_key), timeout=30
+            f"{self._base_url}/export", headers=get_headers(self._api_key), timeout=30
         )
         if resp.status_code >= 400:
             raise AgentXExportError(f"Export manifest failed ({resp.status_code}): {resp.text[:200]}")
@@ -54,7 +56,7 @@ class ExportClient:
         timestamp column, e.g. ``createdAt`` for traces, ``lastSeenAt`` for signals)."""
         params = {"since": since} if since else None
         resp = requests.get(
-            f"{api_base()}/export/{entity}",
+            f"{self._base_url}/export/{entity}",
             headers=get_headers(self._api_key),
             params=params,
             stream=True,

{agentx_python-0.6.34 → agentx_python-0.6.36}/agentx/feedback.py

@@ -26,8 +26,10 @@ class FeedbackClient:
     reopened", reported by a workflow); feedback is a human vote with up/down semantics.
     """
 
-    def __init__(self, api_key: Optional[str] = None):
+    def __init__(self, api_key: Optional[str] = None, base_url: Optional[str] = None):
         self._api_key = api_key
+        # Captured once at construction (deep-dive round 3, bug #1).
+        self._base_url = (base_url or api_base()).rstrip("/")
 
     def report(
         self,
@@ -61,7 +63,7 @@ class FeedbackClient:
             payload["endUserId"] = end_user_id
 
         resp = requests.post(
-            f"{api_base()}/feedback",
+            f"{self._base_url}/feedback",
             headers={**get_headers(self._api_key), "Content-Type": "application/json"},
             json=payload,
             timeout=10,
@@ -79,7 +81,7 @@ class FeedbackClient:
         """All votes recorded on one trace, oldest first (``GET /feedback/trace/:traceId``) -
         the same rows the dashboard's trace dialog shows as up/down chips."""
         resp = requests.get(
-            f"{api_base()}/feedback/trace/{trace_id}",
+            f"{self._base_url}/feedback/trace/{trace_id}",
             headers=get_headers(self._api_key),
             timeout=10,
         )

{agentx_python-0.6.34 → agentx_python-0.6.36}/agentx/monitor/client.py

@@ -85,9 +85,18 @@ class MonitorClient:
         from agentx.monitor.scorers import ScorersClient
         # Scorers-catalog administration as code: template enable/disable, code/external scorer
         # CRUD and dry runs - full parity with the dashboard's Scorers page (P1.3).
-        self.scorers = ScorersClient(api_key=api_key)
+        # Handed this client's own resolved API root, never the process-global default, so a
+        # second AgentX() with a different base_url can't re-point it (deep-dive bug #1).
+        self.scorers = ScorersClient(api_key=api_key, base_url=self._api_root())
+        from agentx.monitor.judge_scorers import JudgeScorersClient
+        # The unified LLM Judge Scorer (rubric + offline/online profiles in one entity) - the
+        # surface that matches the product; evaluations.settings and online_evaluators below
+        # remain as its profile-level views.
+        self.judge_scorers = JudgeScorersClient(api_key=api_key, base_url=self._api_root())
         self.profile = MonitorProfileClient(self)
-        self.online_evaluators = MonitorOnlineEvaluatorClient(self)
+        # Legacy view of an LLM Judge Scorer's online profile - constructed lazily so its
+        # DeprecationWarning fires on first USE, not for every client that never touches it.
+        self._online_evaluators: "MonitorOnlineEvaluatorClient | None" = None
         from agentx.monitor.sessions import MonitorSessionClient
         from agentx.monitor.agents import MonitorAgentClient
         self.sessions = MonitorSessionClient(self)
@@ -171,6 +180,14 @@ class MonitorClient:
         data = self._request("POST", "/online-evaluators", json=self._with_workspace(payload))
         return MonitorOnlineEvaluator(**data["evaluator"])
 
+    @property
+    def online_evaluators(self) -> "MonitorOnlineEvaluatorClient":
+        if self._online_evaluators is None:
+            from agentx.monitor.online_evaluators import MonitorOnlineEvaluatorClient
+
+            self._online_evaluators = MonitorOnlineEvaluatorClient(self)
+        return self._online_evaluators
+
     def list_online_evaluators(self) -> List[MonitorOnlineEvaluator]:
         data = self._request("GET", "/online-evaluators", params=self._workspace_params())
         return [MonitorOnlineEvaluator(**e) for e in data.get("evaluators", [])]