unique-user-memory 2026.34.0.dev4__tar.gz → 2026.34.0.dev6__tar.gz

{unique_user_memory-2026.34.0.dev4 → unique_user_memory-2026.34.0.dev6}/PKG-INFO

@@ -1,14 +1,14 @@
 Metadata-Version: 2.3
 Name: unique-user-memory
-Version: 2026.34.0.dev4
+Version: 2026.34.0.dev6
 Summary: 
 Author: Fabian Schläpfer
 Author-email: Fabian Schläpfer <fabian@unique.ch>
 License: Proprietary
 Requires-Dist: jinja2>=3.1.6
 Requires-Dist: pydantic>=2.8.2
-Requires-Dist: unique-sdk>=2026.34.0.dev5,<2026.34.0rc0
-Requires-Dist: unique-toolkit>=2026.34.0.dev11,<2026.34.0rc0
+Requires-Dist: unique-sdk>=2026.34.0.dev7,<2026.34.0rc0
+Requires-Dist: unique-toolkit>=2026.34.0.dev17,<2026.34.0rc0
 Requires-Python: >=3.12, <4
 Description-Content-Type: text/markdown
 

{unique_user_memory-2026.34.0.dev4 → unique_user_memory-2026.34.0.dev6}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "unique_user_memory"
-version = "2026.34.0.dev4"
+version = "2026.34.0.dev6"
 description = ""
 authors = [
     { name = "Fabian Schläpfer", email = "fabian@unique.ch" },
@@ -11,8 +11,8 @@ requires-python = ">=3.12,<4"
 dependencies = [
     "jinja2>=3.1.6",
     "pydantic>=2.8.2",
-    "unique-sdk>=2026.34.0.dev5,<2026.34.0rc0",
-    "unique-toolkit>=2026.34.0.dev11,<2026.34.0rc0",
+    "unique-sdk>=2026.34.0.dev7,<2026.34.0rc0",
+    "unique-toolkit>=2026.34.0.dev17,<2026.34.0rc0",
 ]
 
 [dependency-groups]

{unique_user_memory-2026.34.0.dev4 → unique_user_memory-2026.34.0.dev6}/unique_user_memory/config.py

@@ -28,7 +28,7 @@ class UserMemoryConfig(BaseModel):
             "'Use Orchestrator Language Model' is false."
         ),
     )
-    max_tokens: int = Field(
+    max_tokens: Annotated[int, RJSFMetaTag.SpecialWidget.hidden()] = Field(
         default=2000,
         ge=500,
         le=8000,

{unique_user_memory-2026.34.0.dev4 → unique_user_memory-2026.34.0.dev6}/unique_user_memory/tests/test_user_memory.py

@@ -13,7 +13,7 @@ from unique_toolkit.chat.schemas import MessageLogStatus
 from unique_toolkit.language_model.default_language_model import (
     DEFAULT_LANGUAGE_MODEL,
 )
-from unique_toolkit.language_model.infos import LanguageModelInfo
+from unique_toolkit.language_model.infos import LanguageModelInfo, LanguageModelName
 from unique_toolkit.language_model.invocation_stats import LanguageModelInvocationStats
 from unique_toolkit.language_model.schemas import (
     LanguageModelAssistantMessage,
@@ -25,6 +25,7 @@ from unique_toolkit.language_model.schemas import (
 from unique_user_memory.config import UserMemoryConfig
 from unique_user_memory.user_memory import (
     UserMemoryState,
+    _gate_max_tokens,
     _sanitize_for_xml_context,
     condense_user_memory,
     consolidate_user_memory,
@@ -621,6 +622,47 @@ async def test_should_consolidate_memory_returns_false_on_noop(
     )
 
 
+def test_gate_max_tokens_depends_on_reasoning_capability() -> None:
+    reasoning_model = LanguageModelInfo.from_name(
+        LanguageModelName.AZURE_GPT_5_2025_0807
+    )
+
+    assert _gate_max_tokens(language_model=_TEST_LANGUAGE_MODEL) == 4
+    assert _gate_max_tokens(language_model=reasoning_model) == 64
+
+
+@pytest.mark.asyncio
+async def test_should_consolidate_memory_uses_reasoning_safe_gate_budget(
+    monkeypatch: pytest.MonkeyPatch,
+) -> None:
+    response = MagicMock()
+    response.choices[0].message.content = "NOOP"
+    llm_service = MagicMock()
+    llm_service.complete_async = AsyncMock(return_value=response)
+    monkeypatch.setattr(
+        "unique_user_memory.user_memory.LanguageModelService",
+        MagicMock(return_value=llm_service),
+    )
+    reasoning_model = LanguageModelInfo.from_name(
+        LanguageModelName.AZURE_GPT_5_2025_0807
+    )
+
+    result = await should_consolidate_memory(
+        current_memory=empty_profile("user_1"),
+        user_id="user_1",
+        user_message="hello",
+        assistant_message="hi",
+        language_model=reasoning_model,
+        event=MagicMock(),
+        logger=MagicMock(),
+    )
+
+    assert result is False
+    assert (
+        llm_service.complete_async.call_args.kwargs["other_options"]["max_tokens"] == 64
+    )
+
+
 @pytest.mark.asyncio
 async def test_should_consolidate_memory_returns_true_on_update(
     monkeypatch: pytest.MonkeyPatch,

{unique_user_memory-2026.34.0.dev4 → unique_user_memory-2026.34.0.dev6}/unique_user_memory/user_memory.py

@@ -20,7 +20,10 @@ from unique_toolkit.language_model import (
     TypeDecoder,
     TypeEncoder,
 )
-from unique_toolkit.language_model.infos import LanguageModelInfo
+from unique_toolkit.language_model.infos import (
+    LanguageModelInfo,
+    ModelCapabilities,
+)
 from unique_toolkit.language_model.invocation_stats import (
     LanguageModelInvocationStats,
 )
@@ -54,6 +57,11 @@ async def noop_update_callback() -> None:
 # The gate only ever replies with the single word UPDATE or NOOP; a tiny
 # output budget keeps the common (NOOP) path cheap and fast.
 _GATE_MAX_TOKENS = 4
+# Reasoning models spend `max_completion_tokens` on internal reasoning before
+# emitting the answer; a 4-token budget truncates the reply to empty and the
+# gate silently degrades to always-UPDATE. Give them enough room to reason and
+# still emit the single-word answer.
+_GATE_MAX_TOKENS_REASONING = 64
 # When condensing an oversized profile, aim below the hard cap so the LLM
 # output leaves headroom and the hard-cut safety net rarely has to fire.
 _CONDENSE_TARGET_RATIO = 0.9
@@ -110,6 +118,12 @@ class UserMemoryState:
     load_invocation_stats: tuple[LanguageModelInvocationStats, ...] = ()
 
 
+def _gate_max_tokens(*, language_model: LanguageModelInfo) -> int:
+    if ModelCapabilities.REASONING in language_model.capabilities:
+        return _GATE_MAX_TOKENS_REASONING
+    return _GATE_MAX_TOKENS
+
+
 def _get_model_tokenizer(
     *,
     language_model: LanguageModelInfo,
@@ -684,7 +698,9 @@ async def should_consolidate_memory(
         response = await llm_service.complete_async(
             messages=messages,
             model_name=language_model.name,
-            other_options={"max_tokens": _GATE_MAX_TOKENS},
+            other_options={
+                "max_tokens": _gate_max_tokens(language_model=language_model)
+            },
         )
     except Exception as exc:
         logger.warning(