agentx-python 0.8.11__tar.gz → 0.8.13__tar.gz
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- {agentx_python-0.8.11 → agentx_python-0.8.13}/PKG-INFO +7 -2
- {agentx_python-0.8.11 → agentx_python-0.8.13}/README.md +6 -1
- {agentx_python-0.8.11 → agentx_python-0.8.13}/agentx/integrations/_traced_call.py +3 -0
- {agentx_python-0.8.11 → agentx_python-0.8.13}/agentx/monitor/client.py +22 -0
- agentx_python-0.8.13/agentx/tracing/framework_detect.py +64 -0
- {agentx_python-0.8.11 → agentx_python-0.8.13}/agentx/tracing/ingest_client.py +9 -0
- {agentx_python-0.8.11 → agentx_python-0.8.13}/agentx/tracing/tracer.py +39 -9
- {agentx_python-0.8.11 → agentx_python-0.8.13}/agentx/version.py +2 -2
- {agentx_python-0.8.11 → agentx_python-0.8.13}/agentx_python.egg-info/PKG-INFO +7 -2
- {agentx_python-0.8.11 → agentx_python-0.8.13}/agentx_python.egg-info/SOURCES.txt +2 -0
- agentx_python-0.8.13/tests/test_framework_detect.py +125 -0
- {agentx_python-0.8.11 → agentx_python-0.8.13}/LICENSE +0 -0
- {agentx_python-0.8.11 → agentx_python-0.8.13}/agentx/__init__.py +0 -0
- {agentx_python-0.8.11 → agentx_python-0.8.13}/agentx/agentx.py +0 -0
- {agentx_python-0.8.11 → agentx_python-0.8.13}/agentx/cli.py +0 -0
- {agentx_python-0.8.11 → agentx_python-0.8.13}/agentx/evaluations/__init__.py +0 -0
- {agentx_python-0.8.11 → agentx_python-0.8.13}/agentx/evaluations/_term.py +0 -0
- {agentx_python-0.8.11 → agentx_python-0.8.13}/agentx/evaluations/adapters/__init__.py +0 -0
- {agentx_python-0.8.11 → agentx_python-0.8.13}/agentx/evaluations/adapters/http_endpoint.py +0 -0
- {agentx_python-0.8.11 → agentx_python-0.8.13}/agentx/evaluations/adapters/precomputed.py +0 -0
- {agentx_python-0.8.11 → agentx_python-0.8.13}/agentx/evaluations/adapters/raw.py +0 -0
- {agentx_python-0.8.11 → agentx_python-0.8.13}/agentx/evaluations/client.py +0 -0
- {agentx_python-0.8.11 → agentx_python-0.8.13}/agentx/evaluations/datasets.py +0 -0
- {agentx_python-0.8.11 → agentx_python-0.8.13}/agentx/evaluations/evaluation_settings.py +0 -0
- {agentx_python-0.8.11 → agentx_python-0.8.13}/agentx/evaluations/models.py +0 -0
- {agentx_python-0.8.11 → agentx_python-0.8.13}/agentx/evaluations/prompts.py +0 -0
- {agentx_python-0.8.11 → agentx_python-0.8.13}/agentx/evaluations/reporting.py +0 -0
- {agentx_python-0.8.11 → agentx_python-0.8.13}/agentx/evaluations/results.py +0 -0
- {agentx_python-0.8.11 → agentx_python-0.8.13}/agentx/evaluations/runner.py +0 -0
- {agentx_python-0.8.11 → agentx_python-0.8.13}/agentx/evaluations/tool_schemas.py +0 -0
- {agentx_python-0.8.11 → agentx_python-0.8.13}/agentx/evaluations/tracing.py +0 -0
- {agentx_python-0.8.11 → agentx_python-0.8.13}/agentx/exceptions.py +0 -0
- {agentx_python-0.8.11 → agentx_python-0.8.13}/agentx/export.py +0 -0
- {agentx_python-0.8.11 → agentx_python-0.8.13}/agentx/feedback.py +0 -0
- {agentx_python-0.8.11 → agentx_python-0.8.13}/agentx/integrations/__init__.py +0 -0
- {agentx_python-0.8.11 → agentx_python-0.8.13}/agentx/integrations/anthropic.py +0 -0
- {agentx_python-0.8.11 → agentx_python-0.8.13}/agentx/integrations/autogen.py +0 -0
- {agentx_python-0.8.11 → agentx_python-0.8.13}/agentx/integrations/crewai.py +0 -0
- {agentx_python-0.8.11 → agentx_python-0.8.13}/agentx/integrations/databricks.py +0 -0
- {agentx_python-0.8.11 → agentx_python-0.8.13}/agentx/integrations/google_adk.py +0 -0
- {agentx_python-0.8.11 → agentx_python-0.8.13}/agentx/integrations/google_genai.py +0 -0
- {agentx_python-0.8.11 → agentx_python-0.8.13}/agentx/integrations/langchain.py +0 -0
- {agentx_python-0.8.11 → agentx_python-0.8.13}/agentx/integrations/litellm.py +0 -0
- {agentx_python-0.8.11 → agentx_python-0.8.13}/agentx/integrations/llamaindex.py +0 -0
- {agentx_python-0.8.11 → agentx_python-0.8.13}/agentx/integrations/moveworks.py +0 -0
- {agentx_python-0.8.11 → agentx_python-0.8.13}/agentx/integrations/openai.py +0 -0
- {agentx_python-0.8.11 → agentx_python-0.8.13}/agentx/integrations/openai_agents.py +0 -0
- {agentx_python-0.8.11 → agentx_python-0.8.13}/agentx/monitor/__init__.py +0 -0
- {agentx_python-0.8.11 → agentx_python-0.8.13}/agentx/monitor/agents.py +0 -0
- {agentx_python-0.8.11 → agentx_python-0.8.13}/agentx/monitor/judge_scorers.py +0 -0
- {agentx_python-0.8.11 → agentx_python-0.8.13}/agentx/monitor/models.py +0 -0
- {agentx_python-0.8.11 → agentx_python-0.8.13}/agentx/monitor/online_evaluators.py +0 -0
- {agentx_python-0.8.11 → agentx_python-0.8.13}/agentx/monitor/patterns.py +0 -0
- {agentx_python-0.8.11 → agentx_python-0.8.13}/agentx/monitor/profile.py +0 -0
- {agentx_python-0.8.11 → agentx_python-0.8.13}/agentx/monitor/review_queue.py +0 -0
- {agentx_python-0.8.11 → agentx_python-0.8.13}/agentx/monitor/rules.py +0 -0
- {agentx_python-0.8.11 → agentx_python-0.8.13}/agentx/monitor/scorers.py +0 -0
- {agentx_python-0.8.11 → agentx_python-0.8.13}/agentx/monitor/sessions.py +0 -0
- {agentx_python-0.8.11 → agentx_python-0.8.13}/agentx/monitor/signals.py +0 -0
- {agentx_python-0.8.11 → agentx_python-0.8.13}/agentx/outcomes.py +0 -0
- {agentx_python-0.8.11 → agentx_python-0.8.13}/agentx/projects.py +0 -0
- {agentx_python-0.8.11 → agentx_python-0.8.13}/agentx/py.typed +0 -0
- {agentx_python-0.8.11 → agentx_python-0.8.13}/agentx/resources/__init__.py +0 -0
- {agentx_python-0.8.11 → agentx_python-0.8.13}/agentx/resources/agent.py +0 -0
- {agentx_python-0.8.11 → agentx_python-0.8.13}/agentx/resources/conversation.py +0 -0
- {agentx_python-0.8.11 → agentx_python-0.8.13}/agentx/resources/workforce.py +0 -0
- {agentx_python-0.8.11 → agentx_python-0.8.13}/agentx/testing.py +0 -0
- {agentx_python-0.8.11 → agentx_python-0.8.13}/agentx/traces.py +0 -0
- {agentx_python-0.8.11 → agentx_python-0.8.13}/agentx/tracing/__init__.py +0 -0
- {agentx_python-0.8.11 → agentx_python-0.8.13}/agentx/tracing/ci_types.py +0 -0
- {agentx_python-0.8.11 → agentx_python-0.8.13}/agentx/tracing/eval_scope.py +0 -0
- {agentx_python-0.8.11 → agentx_python-0.8.13}/agentx/util.py +0 -0
- {agentx_python-0.8.11 → agentx_python-0.8.13}/agentx_python.egg-info/dependency_links.txt +0 -0
- {agentx_python-0.8.11 → agentx_python-0.8.13}/agentx_python.egg-info/entry_points.txt +0 -0
- {agentx_python-0.8.11 → agentx_python-0.8.13}/agentx_python.egg-info/not-zip-safe +0 -0
- {agentx_python-0.8.11 → agentx_python-0.8.13}/agentx_python.egg-info/requires.txt +0 -0
- {agentx_python-0.8.11 → agentx_python-0.8.13}/agentx_python.egg-info/top_level.txt +0 -0
- {agentx_python-0.8.11 → agentx_python-0.8.13}/setup.cfg +0 -0
- {agentx_python-0.8.11 → agentx_python-0.8.13}/setup.py +0 -0
- {agentx_python-0.8.11 → agentx_python-0.8.13}/tests/test_cli_launcher.py +0 -0
- {agentx_python-0.8.11 → agentx_python-0.8.13}/tests/test_deep_dive_fixes.py +0 -0
- {agentx_python-0.8.11 → agentx_python-0.8.13}/tests/test_docs_match_sdk.py +0 -0
- {agentx_python-0.8.11 → agentx_python-0.8.13}/tests/test_eval_scope.py +0 -0
- {agentx_python-0.8.11 → agentx_python-0.8.13}/tests/test_integration.py +0 -0
- {agentx_python-0.8.11 → agentx_python-0.8.13}/tests/test_integrations.py +0 -0
- {agentx_python-0.8.11 → agentx_python-0.8.13}/tests/test_judge_scorers.py +0 -0
- {agentx_python-0.8.11 → agentx_python-0.8.13}/tests/test_pairwise.py +0 -0
- {agentx_python-0.8.11 → agentx_python-0.8.13}/tests/test_review_queue.py +0 -0
- {agentx_python-0.8.11 → agentx_python-0.8.13}/tests/test_runner_features.py +0 -0
- {agentx_python-0.8.11 → agentx_python-0.8.13}/tests/test_selfhost_analysis_fallback.py +0 -0
- {agentx_python-0.8.11 → agentx_python-0.8.13}/tests/test_span_tree.py +0 -0
- {agentx_python-0.8.11 → agentx_python-0.8.13}/tests/test_testing.py +0 -0
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
Metadata-Version: 2.4
|
|
2
2
|
Name: agentx-python
|
|
3
|
-
Version: 0.8.
|
|
3
|
+
Version: 0.8.13
|
|
4
4
|
Summary: Official Python SDK for AgentX (https://www.agentx.so/)
|
|
5
5
|
Home-page: https://github.com/AgentX-ai/AgentX-python
|
|
6
6
|
Author: Robin Wang and AgentX Team
|
|
@@ -252,7 +252,12 @@ extra:
|
|
|
252
252
|
| LlamaIndex | `pip install "agentx-python[llamaindex]"` | `AgentXLlamaIndexHandler` |
|
|
253
253
|
| AutoGen | `pip install "agentx-python[autogen]"` | `AgentXAutoGenObserver` |
|
|
254
254
|
|
|
255
|
-
Or plain Python - wrap any function with `@tracer.trace(...)` and it just works, no framework
|
|
255
|
+
Or plain Python - wrap any function with `@tracer.trace(...)` and it just works, no framework
|
|
256
|
+
required. Tracing is **platform agnostic**: each integration stamps its platform label
|
|
257
|
+
automatically, a plain trace auto-detects the one orchestration framework imported in the
|
|
258
|
+
process, and `framework="any-name"` labels platforms AgentX has never heard of - the label
|
|
259
|
+
drives the dashboard's framework filter and Monitor's Platforms chart. See
|
|
260
|
+
[Platform detection](TRACING.md#platform-detection).
|
|
256
261
|
|
|
257
262
|
Running specialist agents in parallel with a `ThreadPoolExecutor`? Wrap each worker body in `tracer.use_span(span)` so their steps land on the parent trace instead of becoming independent traces - see [TRACING.md](TRACING.md) for the full pattern.
|
|
258
263
|
|
|
@@ -188,7 +188,12 @@ extra:
|
|
|
188
188
|
| LlamaIndex | `pip install "agentx-python[llamaindex]"` | `AgentXLlamaIndexHandler` |
|
|
189
189
|
| AutoGen | `pip install "agentx-python[autogen]"` | `AgentXAutoGenObserver` |
|
|
190
190
|
|
|
191
|
-
Or plain Python - wrap any function with `@tracer.trace(...)` and it just works, no framework
|
|
191
|
+
Or plain Python - wrap any function with `@tracer.trace(...)` and it just works, no framework
|
|
192
|
+
required. Tracing is **platform agnostic**: each integration stamps its platform label
|
|
193
|
+
automatically, a plain trace auto-detects the one orchestration framework imported in the
|
|
194
|
+
process, and `framework="any-name"` labels platforms AgentX has never heard of - the label
|
|
195
|
+
drives the dashboard's framework filter and Monitor's Platforms chart. See
|
|
196
|
+
[Platform detection](TRACING.md#platform-detection).
|
|
192
197
|
|
|
193
198
|
Running specialist agents in parallel with a `ThreadPoolExecutor`? Wrap each worker body in `tracer.use_span(span)` so their steps land on the parent trace instead of becoming independent traces - see [TRACING.md](TRACING.md) for the full pattern.
|
|
194
199
|
|
|
@@ -143,6 +143,9 @@ def finish_llm_call(
|
|
|
143
143
|
input=input_repr,
|
|
144
144
|
output=output,
|
|
145
145
|
model=model,
|
|
146
|
+
# Stamp the provider literal on a span opened without one (adoption keeps an
|
|
147
|
+
# explicit framework= or a framework integration's label winning over this).
|
|
148
|
+
framework=framework,
|
|
146
149
|
input_tokens=input_tokens,
|
|
147
150
|
output_tokens=output_tokens,
|
|
148
151
|
cache_read_tokens=cache_read_tokens,
|
|
@@ -268,6 +268,28 @@ class MonitorClient:
|
|
|
268
268
|
plus deltas vs the prior window and the run-outcome breakdown."""
|
|
269
269
|
return self._request("GET", "/kpis", params={"window": window})
|
|
270
270
|
|
|
271
|
+
def metrics(
|
|
272
|
+
self,
|
|
273
|
+
window: str = "1d",
|
|
274
|
+
*,
|
|
275
|
+
agent: Optional[str] = None,
|
|
276
|
+
model: Optional[str] = None,
|
|
277
|
+
tool: Optional[str] = None,
|
|
278
|
+
framework: Optional[str] = None,
|
|
279
|
+
status: Optional[str] = None,
|
|
280
|
+
) -> dict:
|
|
281
|
+
"""The Monitor metrics grid's data over a window ("1h".."90d") - bucketed spans by kind,
|
|
282
|
+
latency percentiles, tokens/cost, tool executions and failures, and platform attribution
|
|
283
|
+
(`frameworks` window totals + per-bucket `byFramework` - the Platforms chart). Optional
|
|
284
|
+
filters scope every number the way the dashboard's filter chips do; `framework` matches
|
|
285
|
+
the platform label traces carry (see TRACING.md's Platform detection), with "other"
|
|
286
|
+
selecting unlabeled traffic."""
|
|
287
|
+
params = {"window": window}
|
|
288
|
+
for key, value in (("agent", agent), ("model", model), ("tool", tool), ("framework", framework), ("status", status)):
|
|
289
|
+
if value is not None:
|
|
290
|
+
params[key] = value
|
|
291
|
+
return self._request("GET", "/metrics", params=params)
|
|
292
|
+
|
|
271
293
|
def topics(self, window: str = "7d") -> dict:
|
|
272
294
|
"""The Topics view's data over a window ("24h", "7d", "30d"): LLM-classified themes of
|
|
273
295
|
sampled production traffic with per-topic counts and sentiment. Empty until Topics is
|
|
@@ -0,0 +1,64 @@
|
|
|
1
|
+
"""Best-effort agent-framework auto-detection (the "platform agnostic" story).
|
|
2
|
+
|
|
3
|
+
A span whose framework was neither passed explicitly (``tracer.trace(...,
|
|
4
|
+
framework="...")``) nor adopted from a framework integration (callback handler,
|
|
5
|
+
observer, patched client - see ``_TraceSpan._captured_framework``) gets labeled
|
|
6
|
+
by looking at which known ORCHESTRATION framework is actually imported in this
|
|
7
|
+
process. ``sys.modules`` is the signal - imported, not merely installed - so a
|
|
8
|
+
machine with ten frameworks pip-installed but one in use still resolves.
|
|
9
|
+
|
|
10
|
+
Only unambiguous answers are given: zero or more than one known framework
|
|
11
|
+
loaded means ``None``, and the span goes out unlabeled rather than mislabeled.
|
|
12
|
+
The user's explicit ``framework=`` always wins, including totally custom names
|
|
13
|
+
for platforms this table has never heard of.
|
|
14
|
+
|
|
15
|
+
Raw provider SDKs (openai, anthropic, google-genai, ...) are deliberately NOT
|
|
16
|
+
in this table: they are transitive dependencies of nearly every framework, so
|
|
17
|
+
their presence says nothing about what orchestrates the agent - and their
|
|
18
|
+
patched-client integrations already stamp the provider literal on the spans
|
|
19
|
+
they create.
|
|
20
|
+
"""
|
|
21
|
+
|
|
22
|
+
from __future__ import annotations
|
|
23
|
+
|
|
24
|
+
import sys
|
|
25
|
+
from typing import Optional
|
|
26
|
+
|
|
27
|
+
# Top-level module name -> the wire literal the matching integration emits.
|
|
28
|
+
# Multiple modules may map to one literal (langgraph is the LangChain family).
|
|
29
|
+
_ORCHESTRATOR_MODULES = {
|
|
30
|
+
"langchain": "langchain",
|
|
31
|
+
"langchain_core": "langchain",
|
|
32
|
+
"langgraph": "langchain",
|
|
33
|
+
"crewai": "crewai",
|
|
34
|
+
"llama_index": "llamaindex",
|
|
35
|
+
"autogen": "autogen",
|
|
36
|
+
"autogen_agentchat": "autogen",
|
|
37
|
+
"agents": "openai-agents", # the OpenAI Agents SDK's import name
|
|
38
|
+
"google.adk": "google-adk",
|
|
39
|
+
"semantic_kernel": "semantic-kernel",
|
|
40
|
+
"haystack": "haystack",
|
|
41
|
+
"pydantic_ai": "pydantic-ai",
|
|
42
|
+
"smolagents": "smolagents",
|
|
43
|
+
"dspy": "dspy",
|
|
44
|
+
}
|
|
45
|
+
|
|
46
|
+
|
|
47
|
+
def _looks_like_openai_agents_sdk() -> bool:
|
|
48
|
+
# "agents" is a name any user package could claim - only trust it when the
|
|
49
|
+
# OpenAI Agents SDK's own submodules are loaded alongside it.
|
|
50
|
+
return "agents.run" in sys.modules or "agents.tracing" in sys.modules
|
|
51
|
+
|
|
52
|
+
|
|
53
|
+
def detect_framework() -> Optional[str]:
|
|
54
|
+
"""The single unambiguous orchestration framework imported right now, else None."""
|
|
55
|
+
found: set = set()
|
|
56
|
+
for module, literal in _ORCHESTRATOR_MODULES.items():
|
|
57
|
+
if module not in sys.modules:
|
|
58
|
+
continue
|
|
59
|
+
if module == "agents" and not _looks_like_openai_agents_sdk():
|
|
60
|
+
continue
|
|
61
|
+
found.add(literal)
|
|
62
|
+
if len(found) > 1:
|
|
63
|
+
return None
|
|
64
|
+
return found.pop() if len(found) == 1 else None
|
|
@@ -428,6 +428,15 @@ class IngestClient:
|
|
|
428
428
|
continue
|
|
429
429
|
|
|
430
430
|
if resp.status_code in _RETRYABLE_STATUS and attempt < _MAX_RETRIES - 1:
|
|
431
|
+
# 429 = the engine's bounded ingest queue shedding load (its ADR-0005): honor
|
|
432
|
+
# Retry-After exactly instead of the generic backoff schedule, so the SDK backs
|
|
433
|
+
# off in step with the server's own flush cadence.
|
|
434
|
+
retry_after = resp.headers.get("Retry-After")
|
|
435
|
+
if resp.status_code == 429 and retry_after:
|
|
436
|
+
try:
|
|
437
|
+
time.sleep(min(30.0, float(retry_after)))
|
|
438
|
+
except ValueError:
|
|
439
|
+
pass
|
|
431
440
|
last_exc = Exception(f"HTTP {resp.status_code}")
|
|
432
441
|
continue
|
|
433
442
|
if not resp.ok:
|
|
@@ -14,6 +14,7 @@ from agentx.exceptions import CIGateFailure
|
|
|
14
14
|
from agentx.tracing.ingest_client import IngestClient
|
|
15
15
|
from agentx.tracing.ci_types import CIRun, CIRunResult, CIRunStatus, CIQuestionScore
|
|
16
16
|
from agentx.tracing.eval_scope import EVAL_RUN_SOURCE, current_eval_run_id
|
|
17
|
+
from agentx.tracing.framework_detect import detect_framework
|
|
17
18
|
|
|
18
19
|
F = TypeVar("F", bound=Callable[..., Any])
|
|
19
20
|
|
|
@@ -125,6 +126,10 @@ class _TraceSpan:
|
|
|
125
126
|
# Adopted from a merged child run (e.g. AgentXCallbackHandler) when this span itself
|
|
126
127
|
# wasn't opened with an explicit framework= - see _merge_child_run below.
|
|
127
128
|
self._captured_framework: Optional[str] = None
|
|
129
|
+
# Best-effort auto-detection (framework_detect.py), resolved once at span open so child
|
|
130
|
+
# spans emitted mid-flight carry the same label the root will. Lowest precedence:
|
|
131
|
+
# explicit framework= > integration adoption > this. None when ambiguous.
|
|
132
|
+
self._detected_framework: Optional[str] = None if framework else detect_framework()
|
|
128
133
|
self._input_tokens: int = 0
|
|
129
134
|
self._output_tokens: int = 0
|
|
130
135
|
# Subsets of _input_tokens (not additional tokens) - a prompt-caching write/read, when the
|
|
@@ -191,7 +196,7 @@ class _TraceSpan:
|
|
|
191
196
|
latency_ms=latency_ms,
|
|
192
197
|
error=self._error,
|
|
193
198
|
metadata=metadata,
|
|
194
|
-
framework=self._framework or self._captured_framework,
|
|
199
|
+
framework=self._framework or self._captured_framework or self._detected_framework or detect_framework(),
|
|
195
200
|
model=self._model or self._captured_model,
|
|
196
201
|
tool_calls=self.tool_calls or None,
|
|
197
202
|
session_id=self._session_id,
|
|
@@ -235,14 +240,18 @@ class _TraceSpan:
|
|
|
235
240
|
input: Any = None,
|
|
236
241
|
output: Any = None,
|
|
237
242
|
model: Optional[str] = None,
|
|
243
|
+
framework: Optional[str] = None,
|
|
238
244
|
input_tokens: Optional[int] = None,
|
|
239
245
|
output_tokens: Optional[int] = None,
|
|
240
246
|
cache_read_tokens: Optional[int] = None,
|
|
241
247
|
cache_write_tokens: Optional[int] = None,
|
|
242
248
|
) -> None:
|
|
243
249
|
"""Record one LLM-call child span (e.g. one patched Anthropic call) under this span -
|
|
244
|
-
name left unset so _merge_child_run auto-numbers it "LLM Call N".
|
|
250
|
+
name left unset so _merge_child_run auto-numbers it "LLM Call N". ``framework`` lets the
|
|
251
|
+
patched client stamp its provider literal on a span the user opened without one - the
|
|
252
|
+
adoption in _merge_child_run keeps explicit/integration labels winning."""
|
|
245
253
|
self._merge_child_run(
|
|
254
|
+
framework=framework,
|
|
246
255
|
execution_steps=[{
|
|
247
256
|
"duration_ms": duration_ms,
|
|
248
257
|
"start_time": start_time,
|
|
@@ -301,7 +310,7 @@ class _TraceSpan:
|
|
|
301
310
|
child = _TraceSpan(
|
|
302
311
|
tracer=self._tracer,
|
|
303
312
|
name=name,
|
|
304
|
-
framework=framework or self._framework or self._captured_framework,
|
|
313
|
+
framework=framework or self._framework or self._captured_framework or self._detected_framework,
|
|
305
314
|
model=model,
|
|
306
315
|
session_id=self._session_id,
|
|
307
316
|
)
|
|
@@ -397,6 +406,13 @@ class _TraceSpan:
|
|
|
397
406
|
under this span).
|
|
398
407
|
"""
|
|
399
408
|
with self._merge_lock:
|
|
409
|
+
# Adopt framework/model BEFORE emitting child spans: child_span resolves its
|
|
410
|
+
# framework from this span's fields, so adopting after the emission loops used to
|
|
411
|
+
# send every CrewAI/AutoGen child out unlabeled while only the root got stamped.
|
|
412
|
+
if model and not self._captured_model:
|
|
413
|
+
self._captured_model = model
|
|
414
|
+
if framework and not self._captured_framework:
|
|
415
|
+
self._captured_framework = framework
|
|
400
416
|
for step in [] if not emit_steps else (execution_steps or []):
|
|
401
417
|
self._child_span_count += 1
|
|
402
418
|
self.child_span(
|
|
@@ -463,10 +479,6 @@ class _TraceSpan:
|
|
|
463
479
|
self.input = input
|
|
464
480
|
if output is not None:
|
|
465
481
|
self.output = output
|
|
466
|
-
if model and not self._captured_model:
|
|
467
|
-
self._captured_model = model
|
|
468
|
-
if framework and not self._captured_framework:
|
|
469
|
-
self._captured_framework = framework
|
|
470
482
|
if input_tokens:
|
|
471
483
|
self._input_tokens += input_tokens
|
|
472
484
|
if output_tokens:
|
|
@@ -520,7 +532,8 @@ class _TraceSpan:
|
|
|
520
532
|
# to be called from inside another active span.
|
|
521
533
|
span = self._tracer.trace(
|
|
522
534
|
self.name, metadata=self._metadata, framework=self._framework, model=self._model,
|
|
523
|
-
session_id=self._session_id,
|
|
535
|
+
session_id=self._session_id, sync=self._sync, monitor=self._monitor,
|
|
536
|
+
pattern_ids=self._pattern_ids, agent_id=self._agent_id, span_kind=self._span_kind,
|
|
524
537
|
)
|
|
525
538
|
span.__enter__()
|
|
526
539
|
try:
|
|
@@ -542,7 +555,8 @@ class _TraceSpan:
|
|
|
542
555
|
# See _wrap_sync's comment - same "fresh span per call" reasoning applies here.
|
|
543
556
|
span = self._tracer.trace(
|
|
544
557
|
self.name, metadata=self._metadata, framework=self._framework, model=self._model,
|
|
545
|
-
session_id=self._session_id,
|
|
558
|
+
session_id=self._session_id, sync=self._sync, monitor=self._monitor,
|
|
559
|
+
pattern_ids=self._pattern_ids, agent_id=self._agent_id, span_kind=self._span_kind,
|
|
546
560
|
)
|
|
547
561
|
span.__enter__()
|
|
548
562
|
try:
|
|
@@ -911,6 +925,22 @@ class Tracer:
|
|
|
911
925
|
|
|
912
926
|
with client.tracer.trace("support-agent", agent_id="ag_123", sync=True) as span:
|
|
913
927
|
span.output = call_llm(...)
|
|
928
|
+
|
|
929
|
+
``framework`` names the platform the agent runs on - tracing is platform agnostic, and
|
|
930
|
+
this label is how the dashboard's framework filter and Monitor's Platforms chart group
|
|
931
|
+
traffic. Three ways it gets set, strongest first:
|
|
932
|
+
|
|
933
|
+
1. **Explicit**: ``framework="langchain"`` - any string works, including platforms
|
|
934
|
+
AgentX has no integration for (``framework="my-inhouse-runner"``).
|
|
935
|
+
2. **Integration**: every integration stamps its own literal automatically -
|
|
936
|
+
``langchain``, ``crewai``, ``openai-agents``, ``openai``, ``anthropic``,
|
|
937
|
+
``google-genai``, ``google-adk``, ``litellm``, ``llamaindex``, ``autogen``,
|
|
938
|
+
``moveworks``, ``databricks``.
|
|
939
|
+
3. **Auto-detection**: with neither of the above, the SDK labels the span with the one
|
|
940
|
+
known orchestration framework imported in the process (LangChain/LangGraph, CrewAI,
|
|
941
|
+
LlamaIndex, AutoGen, OpenAI Agents SDK, Google ADK, Semantic Kernel, Haystack,
|
|
942
|
+
Pydantic AI, smolagents, DSPy). Ambiguous (several imported) or unknown -> the span
|
|
943
|
+
goes out unlabeled rather than mislabeled.
|
|
914
944
|
"""
|
|
915
945
|
return _TraceSpan(
|
|
916
946
|
tracer=self,
|
|
@@ -1,7 +1,7 @@
|
|
|
1
|
-
VERSION = "0.8.
|
|
1
|
+
VERSION = "0.8.13"
|
|
2
2
|
|
|
3
3
|
# The AgentX-trace-eval release this SDK version is tested against - what `agentx-trace-eval`
|
|
4
4
|
# installs and converges to (see agentx/cli.py). Bump together with VERSION when releasing, so
|
|
5
5
|
# every published SDK names a known-good engine+dashboard pair. Users can override with
|
|
6
6
|
# AGENTX_TRACE_EVAL_VERSION=<tag|latest>.
|
|
7
|
-
ENGINE_VERSION = "v0.3.
|
|
7
|
+
ENGINE_VERSION = "v0.3.7"
|
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
Metadata-Version: 2.4
|
|
2
2
|
Name: agentx-python
|
|
3
|
-
Version: 0.8.
|
|
3
|
+
Version: 0.8.13
|
|
4
4
|
Summary: Official Python SDK for AgentX (https://www.agentx.so/)
|
|
5
5
|
Home-page: https://github.com/AgentX-ai/AgentX-python
|
|
6
6
|
Author: Robin Wang and AgentX Team
|
|
@@ -252,7 +252,12 @@ extra:
|
|
|
252
252
|
| LlamaIndex | `pip install "agentx-python[llamaindex]"` | `AgentXLlamaIndexHandler` |
|
|
253
253
|
| AutoGen | `pip install "agentx-python[autogen]"` | `AgentXAutoGenObserver` |
|
|
254
254
|
|
|
255
|
-
Or plain Python - wrap any function with `@tracer.trace(...)` and it just works, no framework
|
|
255
|
+
Or plain Python - wrap any function with `@tracer.trace(...)` and it just works, no framework
|
|
256
|
+
required. Tracing is **platform agnostic**: each integration stamps its platform label
|
|
257
|
+
automatically, a plain trace auto-detects the one orchestration framework imported in the
|
|
258
|
+
process, and `framework="any-name"` labels platforms AgentX has never heard of - the label
|
|
259
|
+
drives the dashboard's framework filter and Monitor's Platforms chart. See
|
|
260
|
+
[Platform detection](TRACING.md#platform-detection).
|
|
256
261
|
|
|
257
262
|
Running specialist agents in parallel with a `ThreadPoolExecutor`? Wrap each worker body in `tracer.use_span(span)` so their steps land on the parent trace instead of becoming independent traces - see [TRACING.md](TRACING.md) for the full pattern.
|
|
258
263
|
|
|
@@ -64,6 +64,7 @@ agentx/resources/workforce.py
|
|
|
64
64
|
agentx/tracing/__init__.py
|
|
65
65
|
agentx/tracing/ci_types.py
|
|
66
66
|
agentx/tracing/eval_scope.py
|
|
67
|
+
agentx/tracing/framework_detect.py
|
|
67
68
|
agentx/tracing/ingest_client.py
|
|
68
69
|
agentx/tracing/tracer.py
|
|
69
70
|
agentx_python.egg-info/PKG-INFO
|
|
@@ -77,6 +78,7 @@ tests/test_cli_launcher.py
|
|
|
77
78
|
tests/test_deep_dive_fixes.py
|
|
78
79
|
tests/test_docs_match_sdk.py
|
|
79
80
|
tests/test_eval_scope.py
|
|
81
|
+
tests/test_framework_detect.py
|
|
80
82
|
tests/test_integration.py
|
|
81
83
|
tests/test_integrations.py
|
|
82
84
|
tests/test_judge_scorers.py
|
|
@@ -0,0 +1,125 @@
|
|
|
1
|
+
"""
|
|
2
|
+
Platform-agnostic framework capture: explicit framework= always wins, integrations stamp their
|
|
3
|
+
literal, and - new - the SDK auto-detects the single unambiguous orchestration framework
|
|
4
|
+
imported in the process (agentx/tracing/framework_detect.py). Also pins the two capture-gap
|
|
5
|
+
fixes: _merge_child_run adopts the framework BEFORE emitting child spans (CrewAI/AutoGen
|
|
6
|
+
children used to go out unlabeled), and _record_llm_call forwards the patched client's provider
|
|
7
|
+
literal onto a user-opened span that has no label of its own.
|
|
8
|
+
|
|
9
|
+
Same harness as test_span_tree.py: only the ingest_client boundary is mocked, so the real
|
|
10
|
+
_send/_dispatch/child_span paths run and every wire dict is inspectable.
|
|
11
|
+
"""
|
|
12
|
+
from __future__ import annotations
|
|
13
|
+
|
|
14
|
+
import sys
|
|
15
|
+
import types
|
|
16
|
+
from unittest.mock import MagicMock
|
|
17
|
+
|
|
18
|
+
import pytest
|
|
19
|
+
|
|
20
|
+
from agentx.tracing.framework_detect import _ORCHESTRATOR_MODULES, detect_framework
|
|
21
|
+
from agentx.tracing.tracer import Tracer
|
|
22
|
+
|
|
23
|
+
|
|
24
|
+
def make_tracer() -> Tracer:
|
|
25
|
+
return Tracer(ingest_client=MagicMock())
|
|
26
|
+
|
|
27
|
+
|
|
28
|
+
def enqueued_wires(tracer: Tracer) -> list:
|
|
29
|
+
return [call.args[0] for call in tracer._client.enqueue.call_args_list]
|
|
30
|
+
|
|
31
|
+
|
|
32
|
+
@pytest.fixture()
|
|
33
|
+
def clean_modules(monkeypatch):
|
|
34
|
+
"""Remove every known orchestrator from sys.modules so each test states its own world."""
|
|
35
|
+
for module in list(_ORCHESTRATOR_MODULES) + ["agents.run", "agents.tracing"]:
|
|
36
|
+
monkeypatch.delitem(sys.modules, module, raising=False)
|
|
37
|
+
return monkeypatch
|
|
38
|
+
|
|
39
|
+
|
|
40
|
+
def fake_import(monkeypatch, name: str) -> None:
|
|
41
|
+
monkeypatch.setitem(sys.modules, name, types.ModuleType(name))
|
|
42
|
+
|
|
43
|
+
|
|
44
|
+
def test_detects_single_imported_orchestrator(clean_modules):
|
|
45
|
+
fake_import(clean_modules, "crewai")
|
|
46
|
+
assert detect_framework() == "crewai"
|
|
47
|
+
tracer = make_tracer()
|
|
48
|
+
with tracer.trace("agent"):
|
|
49
|
+
pass
|
|
50
|
+
assert enqueued_wires(tracer)[0]["framework"] == "crewai"
|
|
51
|
+
|
|
52
|
+
|
|
53
|
+
def test_no_orchestrator_means_no_label(clean_modules):
|
|
54
|
+
assert detect_framework() is None
|
|
55
|
+
tracer = make_tracer()
|
|
56
|
+
with tracer.trace("agent"):
|
|
57
|
+
pass
|
|
58
|
+
assert "framework" not in enqueued_wires(tracer)[0]
|
|
59
|
+
|
|
60
|
+
|
|
61
|
+
def test_ambiguous_imports_stay_unlabeled(clean_modules):
|
|
62
|
+
fake_import(clean_modules, "crewai")
|
|
63
|
+
fake_import(clean_modules, "llama_index")
|
|
64
|
+
assert detect_framework() is None
|
|
65
|
+
|
|
66
|
+
|
|
67
|
+
def test_langchain_family_collapses_to_one_literal(clean_modules):
|
|
68
|
+
# langgraph + langchain_core together are ONE framework, not an ambiguity.
|
|
69
|
+
fake_import(clean_modules, "langchain_core")
|
|
70
|
+
fake_import(clean_modules, "langgraph")
|
|
71
|
+
assert detect_framework() == "langchain"
|
|
72
|
+
|
|
73
|
+
|
|
74
|
+
def test_agents_module_needs_sdk_submodules(clean_modules):
|
|
75
|
+
fake_import(clean_modules, "agents") # could be anyone's package named "agents"
|
|
76
|
+
assert detect_framework() is None
|
|
77
|
+
fake_import(clean_modules, "agents.run")
|
|
78
|
+
assert detect_framework() == "openai-agents"
|
|
79
|
+
|
|
80
|
+
|
|
81
|
+
def test_explicit_framework_beats_detection(clean_modules):
|
|
82
|
+
fake_import(clean_modules, "crewai")
|
|
83
|
+
tracer = make_tracer()
|
|
84
|
+
with tracer.trace("agent", framework="my-inhouse-runner"):
|
|
85
|
+
pass
|
|
86
|
+
assert enqueued_wires(tracer)[0]["framework"] == "my-inhouse-runner"
|
|
87
|
+
|
|
88
|
+
|
|
89
|
+
def test_children_inherit_detected_framework(clean_modules):
|
|
90
|
+
fake_import(clean_modules, "llama_index")
|
|
91
|
+
tracer = make_tracer()
|
|
92
|
+
with tracer.trace("agent") as span:
|
|
93
|
+
span.child_span("step", duration_ms=5)
|
|
94
|
+
wires = enqueued_wires(tracer)
|
|
95
|
+
assert [w.get("framework") for w in wires] == ["llamaindex", "llamaindex"]
|
|
96
|
+
|
|
97
|
+
|
|
98
|
+
def test_merge_child_run_adopts_before_emitting_children(clean_modules):
|
|
99
|
+
# The CrewAI/AutoGen shape: root opened with no framework, the merged sub-run carries it.
|
|
100
|
+
# Children must go out labeled too - adoption used to happen after emission.
|
|
101
|
+
tracer = make_tracer()
|
|
102
|
+
with tracer.trace("crew") as span:
|
|
103
|
+
span._merge_child_run(
|
|
104
|
+
framework="crewai",
|
|
105
|
+
execution_steps=[{"duration_ms": 10, "input": "q", "output": "a"}],
|
|
106
|
+
)
|
|
107
|
+
wires = enqueued_wires(tracer)
|
|
108
|
+
assert len(wires) == 2
|
|
109
|
+
assert all(w.get("framework") == "crewai" for w in wires)
|
|
110
|
+
|
|
111
|
+
|
|
112
|
+
def test_record_llm_call_stamps_provider_on_unlabeled_span(clean_modules):
|
|
113
|
+
tracer = make_tracer()
|
|
114
|
+
with tracer.trace("agent") as span:
|
|
115
|
+
span._record_llm_call(duration_ms=7, model="claude-x", framework="anthropic")
|
|
116
|
+
root = enqueued_wires(tracer)[-1]
|
|
117
|
+
assert root["framework"] == "anthropic"
|
|
118
|
+
|
|
119
|
+
|
|
120
|
+
def test_record_llm_call_never_overrides_explicit_framework(clean_modules):
|
|
121
|
+
tracer = make_tracer()
|
|
122
|
+
with tracer.trace("agent", framework="langchain") as span:
|
|
123
|
+
span._record_llm_call(duration_ms=7, model="gpt-x", framework="openai")
|
|
124
|
+
root = enqueued_wires(tracer)[-1]
|
|
125
|
+
assert root["framework"] == "langchain"
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|