Release v0.2.0b1

CHANGELOG.md

@@ -1,22 +1,24 @@
 ## Release History
 
-### Unreleased
+## [0.2.0b1] (2026-06-30)
+
+#### Features Added
+* Raw conversation turns can now be embedded and vector-searched. Set
+  `enable_turn_embeddings=True` (env `ENABLE_TURN_EMBEDDINGS`) to generate an
+  embedding when each turn is written, then call `search_turns()` (sync and
+  async, on both the client and store) to semantically search the raw turn
+  log. See [PR:#22](https://github.com/AzureCosmosDB/AgentMemoryToolkit/pull/22/)
 
 #### Other Changes
 * The memories container's vector index type is now configurable instead of being
   hard-coded to `diskANN`. Set it via the `vector_index_type` argument to
   `create_memory_store(...)` or the `AI_FOUNDRY_EMBEDDING_VECTOR_INDEX_TYPE`
-  environment variable. Allowed values are `diskANN` (default), `quantizedFlat`,
-  and `flat`. This lets the toolkit run against Cosmos DB accounts without the
-  DiskANN capability (for example the classic Cosmos DB emulator), enabling
-  emulator-backed integration test pipelines.
+  environment variable. See [PR:#24](https://github.com/AzureCosmosDB/AgentMemoryToolkit/pull/24)
 * `ai_foundry_endpoint` now accepts a project-scoped Azure AI Foundry URL
   (`https://<resource>.services.ai.azure.com/api/projects/<name>`) in addition
-  to the account-level inference endpoint. The project path is automatically
-  stripped to the inference base, so callers can paste whichever form the
-  Foundry portal shows them without hitting opaque 404s.
+  to the account-level inference endpoint. See [PR:#23](https://github.com/AzureCosmosDB/AgentMemoryToolkit/pull/23)
 
-### 0.1.0b2 (2026-06-03)
+## [0.1.0b2] (2026-06-03)
 
 #### Bugs Fixed
 * Hardened memory extraction: stops emitting phantom/synthesized facts the user never asserted, stops extracting facts from `[assistant]:` turns, stops re-processing already-extracted turns (which previously produced reversed `CONTRADICT` decisions and meta-facts like `"X is contradicted by Y"`), and stops storing near-duplicate episodic memories for the same scope. Episodic memories also now embed the actual content instead of a boilerplate `"intent recorded"` string. See [PR:#20](https://github.com/AzureCosmosDB/AgentMemoryToolkit/pull/20/)

azure.yaml

@@ -2,7 +2,7 @@
 
 name: azure-cosmos-agent-memory
 metadata:
-  template: azure-cosmos-agent-memory@0.1.0b2
+  template: azure-cosmos-agent-memory@0.2.0b1
 
 infra:
   provider: bicep

azure/cosmos/agent_memory/aio/cosmos_memory_client.py

@@ -709,11 +709,12 @@ async def search_turns(
     ) -> list[dict[str, Any]]:
         """Vector-search the raw conversation log (requires turn embeddings).
 
-        Searches the turns container directly. Turns are strictly thread-scoped
-        and only vector-searchable when ``enable_turn_embeddings`` was set when
-        the turns were written. ``user_id`` is required so the search stays
-        within a single partition rather than fanning out across every user's
-        raw turns.
+        Searches the turns container directly. Only vector-searchable when
+        ``enable_turn_embeddings`` was set when the turns were written.
+        ``user_id`` is required and always filters the results. Passing
+        ``thread_id`` as well scopes the search to a single partition; omitting
+        it fans out across partitions, filtered by ``user_id`` in the WHERE
+        clause.
         """
         return await self._get_store().search_turns(
             search_terms=search_terms,

azure/cosmos/agent_memory/aio/store/memory_store.py

@@ -27,6 +27,7 @@
     MemoryConflictError,
     MemoryNotFoundError,
     MemoryTypeMismatchError,
+    ValidationError,
 )
 from azure.cosmos.agent_memory.logging import get_logger
 from azure.cosmos.agent_memory.models import MemoryRecord
@@ -864,6 +865,7 @@ async def search(
     async def search_turns(
         self,
         search_terms: Optional[str] = None,
+        user_id: Optional[str] = None,
         thread_id: Optional[str] = None,
         role: Optional[str] = None,
         hybrid_search: bool = False,
@@ -874,16 +876,19 @@ async def search_turns(
         created_after: Optional[str | datetime] = None,
         created_before: Optional[str | datetime] = None,
         *,
-        user_id: str,
         query: Optional[str] = None,
     ) -> list[dict[str, Any]]:
         """Search raw conversation turns using vector similarity with optional hybrid ranking.
 
-        Turns are strictly thread-scoped and only vector-searchable when turn
-        embeddings were enabled at write time (see ``enable_turn_embeddings``).
-        ``user_id`` is required so the query is scoped to a single partition
-        instead of a cross-partition scan over every user's raw turns.
+        Only vector-searchable when turn embeddings were enabled at write time
+        (see ``enable_turn_embeddings``). ``user_id`` is required and always
+        filters the results. When ``thread_id`` is also supplied the query
+        targets a single partition; when it is omitted the query fans out
+        across partitions and is filtered by ``user_id`` in the WHERE clause.
+
         """
+        if not user_id:
+            raise ValidationError("user_id is required for search_turns")
         terms = require_search_terms(search_terms, query)
         _validate_hybrid_search(hybrid_search, terms)
         top = top_literal(top_k, name="top_k")

azure/cosmos/agent_memory/cosmos_memory_client.py

@@ -672,11 +672,12 @@ def search_turns(
     ) -> list[dict[str, Any]]:
         """Vector-search the raw conversation log (requires turn embeddings).
 
-        Searches the turns container directly. Turns are strictly thread-scoped
-        and only vector-searchable when ``enable_turn_embeddings`` was set when
-        the turns were written. ``user_id`` is required so the search stays
-        within a single partition rather than fanning out across every user's
-        raw turns.
+        Searches the turns container directly. Only vector-searchable when
+        ``enable_turn_embeddings`` was set when the turns were written.
+        ``user_id`` is required and always filters the results. Passing
+        ``thread_id`` as well scopes the search to a single partition; omitting
+        it fans out across partitions, filtered by ``user_id`` in the WHERE
+        clause.
         """
         return self._get_store().search_turns(
             search_terms=search_terms,

azure/cosmos/agent_memory/store/memory_store.py

@@ -25,6 +25,7 @@
     MemoryConflictError,
     MemoryNotFoundError,
     MemoryTypeMismatchError,
+    ValidationError,
 )
 from azure.cosmos.agent_memory.logging import get_logger
 from azure.cosmos.agent_memory.models import MemoryRecord
@@ -901,6 +902,7 @@ def search(
     def search_turns(
         self,
         search_terms: Optional[str] = None,
+        user_id: Optional[str] = None,
         thread_id: Optional[str] = None,
         role: Optional[str] = None,
         hybrid_search: bool = False,
@@ -911,16 +913,19 @@ def search_turns(
         created_after: Optional[str | datetime] = None,
         created_before: Optional[str | datetime] = None,
         *,
-        user_id: str,
         query: Optional[str] = None,
     ) -> list[dict[str, Any]]:
         """Search raw conversation turns using vector similarity with optional hybrid ranking.
 
-        Turns are strictly thread-scoped and only vector-searchable when turn
-        embeddings were enabled at write time (see ``enable_turn_embeddings``).
-        ``user_id`` is required so the query is scoped to a single partition
-        instead of a cross-partition scan over every user's raw turns.
+        Only vector-searchable when turn embeddings were enabled at write time
+        (see ``enable_turn_embeddings``). ``user_id`` is required and always
+        filters the results. When ``thread_id`` is also supplied the query
+        targets a single partition; when it is omitted the query fans out
+        across partitions and is filtered by ``user_id`` in the WHERE clause.
+
         """
+        if not user_id:
+            raise ValidationError("user_id is required for search_turns")
         terms = require_search_terms(search_terms, query)
         _validate_hybrid_search(hybrid_search, terms)
         top = top_literal(top_k, name="top_k")

function_app/requirements.txt

@@ -2,7 +2,7 @@
 azure-functions
 azure-functions-durable
 
-azure-cosmos-agent-memory==0.1.0b2
+azure-cosmos-agent-memory==0.2.0b1
 azure-cosmos>=4.16.0
 azure-identity>=1.20
 

infra/modules/cosmos.bicep

@@ -262,9 +262,6 @@ resource memoriesSummariesContainer 'Microsoft.DocumentDB/databaseAccounts/sqlDa
           }
         ]
         excludedPaths: [
-          {
-            path: '/embedding/?'
-          }
           {
             path: '/source_memory_ids/*'
           }

pyproject.toml

@@ -12,7 +12,7 @@ namespaces = true
 
 [project]
 name = "azure-cosmos-agent-memory"
-version = "0.1.0b2"
+version = "0.2.0b1"
 description = "Store, retrieve, and transform AI agent memories backed by Azure Cosmos DB"
 readme = "README.md"
 license = {file = "LICENSE"}

tests/unit/aio/store/test_memory_store.py

@@ -7,7 +7,7 @@
 
 from azure.cosmos.agent_memory._container_routing import ContainerKey
 from azure.cosmos.agent_memory.aio.store import AsyncMemoryStore
-from azure.cosmos.agent_memory.exceptions import MemoryNotFoundError, MemoryTypeMismatchError
+from azure.cosmos.agent_memory.exceptions import MemoryNotFoundError, MemoryTypeMismatchError, ValidationError
 
 
 class AsyncIterator:
@@ -539,6 +539,52 @@ async def test_search_turns_queries_turns_container():
     assert "VectorDistance(c.embedding, @embedding)" in sql
 
 
+async def test_search_turns_scopes_to_single_partition_with_thread_id():
+    turns = MagicMock()
+    turns.query_items.return_value = AsyncIterator([])
+    embeddings = MagicMock()
+    embeddings.generate = AsyncMock(return_value=[0.1, 0.2])
+    store = AsyncMemoryStore(
+        containers=_containers(turns=turns),
+        embeddings_client=embeddings,
+    )
+
+    await store.search_turns(search_terms="hello", user_id="u1", thread_id="t1")
+
+    kwargs = turns.query_items.call_args.kwargs
+    assert kwargs["partition_key"] == ["u1", "t1"]
+
+
+async def test_search_turns_fans_out_across_partitions_without_thread_id():
+    turns = MagicMock()
+    turns.query_items.return_value = AsyncIterator([])
+    embeddings = MagicMock()
+    embeddings.generate = AsyncMock(return_value=[0.1, 0.2])
+    store = AsyncMemoryStore(
+        containers=_containers(turns=turns),
+        embeddings_client=embeddings,
+    )
+
+    await store.search_turns(search_terms="hello", user_id="u1")
+
+    kwargs = turns.query_items.call_args.kwargs
+    assert "partition_key" not in kwargs
+
+
+async def test_search_turns_requires_user_id():
+    turns = MagicMock()
+    embeddings = MagicMock()
+    embeddings.generate = AsyncMock(return_value=[0.1, 0.2])
+    store = AsyncMemoryStore(
+        containers=_containers(turns=turns),
+        embeddings_client=embeddings,
+    )
+
+    with pytest.raises(ValidationError):
+        await store.search_turns(search_terms="hello", user_id=None)
+    turns.query_items.assert_not_called()
+
+
 async def test_search_does_not_query_turns_container():
     turns = MagicMock()
     turns.query_items.return_value = AsyncIterator([])

tests/unit/store/test_memory_store.py

@@ -6,7 +6,7 @@
 import pytest
 
 from azure.cosmos.agent_memory._container_routing import ContainerKey
-from azure.cosmos.agent_memory.exceptions import MemoryNotFoundError, MemoryTypeMismatchError
+from azure.cosmos.agent_memory.exceptions import MemoryNotFoundError, MemoryTypeMismatchError, ValidationError
 from azure.cosmos.agent_memory.store import MemoryStore
 
 
@@ -539,6 +539,54 @@ def test_search_turns_queries_turns_container():
     assert "VectorDistance(c.embedding, @embedding)" in sql
 
 
+def test_search_turns_scopes_to_single_partition_with_thread_id():
+    turns = MagicMock()
+    turns.query_items.return_value = []
+    embeddings = MagicMock()
+    embeddings.generate.return_value = [0.1, 0.2]
+    store = MemoryStore(
+        containers=_containers(turns=turns),
+        embeddings_client=embeddings,
+    )
+
+    store.search_turns(search_terms="hello", user_id="u1", thread_id="t1")
+
+    kwargs = turns.query_items.call_args.kwargs
+    assert kwargs["partition_key"] == ["u1", "t1"]
+    assert "enable_cross_partition_query" not in kwargs
+
+
+def test_search_turns_fans_out_across_partitions_without_thread_id():
+    turns = MagicMock()
+    turns.query_items.return_value = []
+    embeddings = MagicMock()
+    embeddings.generate.return_value = [0.1, 0.2]
+    store = MemoryStore(
+        containers=_containers(turns=turns),
+        embeddings_client=embeddings,
+    )
+
+    store.search_turns(search_terms="hello", user_id="u1")
+
+    kwargs = turns.query_items.call_args.kwargs
+    assert "partition_key" not in kwargs
+    assert kwargs["enable_cross_partition_query"] is True
+
+
+def test_search_turns_requires_user_id():
+    turns = MagicMock()
+    embeddings = MagicMock()
+    embeddings.generate.return_value = [0.1, 0.2]
+    store = MemoryStore(
+        containers=_containers(turns=turns),
+        embeddings_client=embeddings,
+    )
+
+    with pytest.raises(ValidationError):
+        store.search_turns(search_terms="hello", user_id=None)
+    turns.query_items.assert_not_called()
+
+
 def test_search_does_not_query_turns_container():
     turns = MagicMock()
     turns.query_items.return_value = []