PyPI - power-loop - Versions diffs - 7.1.0__tar.gz → 7.2.0__tar.gz - Mend

@@ -202,6 +202,19 @@ def _maybe_compact_accepts_context(func: Any) -> bool:
 # ── Utility functions (unchanged from old agent.py) ──
+def _llm_endpoint(llm: Any) -> str:
+    """The base URL that serves this LLM (unwrapping retry / proxy wrappers), or ""."""
+    for _ in range(5):
+        if llm is None:
+            break
+        cfg = getattr(llm, "_cfg", None) or getattr(llm, "config", None)
+        url = getattr(cfg, "base_url", None)
+        if url:
+            return str(url)
+        llm = getattr(llm, "inner", None) or getattr(llm, "_inner", None) or getattr(llm, "llm", None)
+    return ""
 def _truncate_result(output: Any, limit: int | None = RESULT_MAX_CHARS) -> str:
     """Cap a tool result before it enters history and the audit log (None = no cap). Keeps the
     head and the tail: errors and summaries are usually at the end."""
@@ -1333,6 +1346,7 @@ class AgentPipeline:
                             success=False, error_type=type(exc).__name__,
                             prompt_tokens=p_est, completion_tokens=c_est,
                             total_tokens=p_est + c_est, estimated=True, outcome=outcome,
+                            endpoint=_llm_endpoint(self.llm),
                         ),
                         round_index=round_index, stream_id="main",
                     )
@@ -1364,6 +1378,9 @@ class AgentPipeline:
                         prompt_cached_tokens=getattr(usage, "prompt_cached_tokens", None),
                         prompt_cache_miss_tokens=getattr(usage, "prompt_cache_miss_tokens", None),
                         estimated=estimated,
+                        endpoint=_llm_endpoint(self.llm),
+                        provider_cost=getattr(usage, "cost", None),
+                        prompt_image_tokens=getattr(usage, "prompt_image_tokens", None),
                     ),
                     round_index=round_index, stream_id="main",
                 )

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: power-loop
-Version: 7.1.0
+Version: 7.2.0
 Summary: Embeddable agent execution kernel — LLM loop, hooks, events, tools, dynamic sub-agents.
 Author-email: zhangran <zhangran24@126.com>
 License: MIT

@@ -15,7 +15,7 @@ Stability tiers
 无版本承诺，可随时变更或删除。
 """
-__version__ = "7.1.0"
+__version__ = "7.2.0"
 # Public LLM contract (SDK-free) re-exported so callers (e.g. writing llm.* hooks or
 # a custom LLMService) don't reach into the internal vendored transport package (H3.4).

@@ -204,6 +204,10 @@ class LLMTokenUsage:
     accepted_tokens: int | None = None
     rejected_tokens: int | None = None
+    #: Charge the provider itself reported for this call (e.g. OpenRouter's ``usage.cost``), in the
+    #: provider's billing currency. ``None`` = not reported; hosts then price tokens themselves.
+    cost: float | None = None
     def as_dict(self) -> dict[str, int | None]:
         return dict(self.__dict__)

@@ -419,6 +419,14 @@ class OpenAICompatibleChatLLMService(LLMService):
         return raw
     def _usage_obj(self, usage: Any = None) -> LLMTokenUsage:
+        def _cost_or_none(v: Any) -> float | None:
+            if isinstance(v, bool) or not isinstance(v, (int, float, str)):
+                return None
+            try:
+                return float(v)
+            except ValueError:
+                return None
         d = self._last_usage if usage is None else self._usage_dict_from_any(usage)
         def _int_or_none(v: Any) -> int | None:
@@ -448,6 +456,7 @@ class OpenAICompatibleChatLLMService(LLMService):
             reasoning_tokens=_int_or_none(d.get("reasoning_tokens")),
             accepted_tokens=_int_or_none(d.get("accepted_tokens")),
             rejected_tokens=_int_or_none(d.get("rejected_tokens")),
+            cost=_cost_or_none(d.get("cost")),
         )
     async def _with_retries(self, fn, *, method: str) -> Any:

@@ -80,6 +80,13 @@ class LlmCallCompletedPayload(BaseEventPayload):
     #: ``ok`` · ``error`` · ``timeout`` · ``aborted`` (the attempt's task was cancelled: a retry
     #: deadline, steering, a stop).
     outcome: str = "ok"
+    #: 7.2: which endpoint served the call (the transport's base URL) — hosts price per connection,
+    #: and one model name can be served by several providers at different prices.
+    endpoint: str = ""
+    #: 7.2: charge the provider reported for this call (``LLMTokenUsage.cost``), ``None`` if not reported.
+    provider_cost: float | None = None
+    #: 7.2: image input tokens, when the provider splits them out (``None`` = not reported).
+    prompt_image_tokens: int | None = None
 @dataclass

power-loop 7.1.0__tar.gz → 7.2.0__tar.gz

power-loop 7.1.0tar.gz → 7.2.0tar.gz