enabled embeddings of raw turns, small other edits by jcodella · Pull Request #22 · AzureCosmosDB/AgentMemoryToolkit

jcodella · 2026-06-22T19:56:18Z

No description provided.

Copilot

Pull request overview

This PR introduces an opt-in switch to embed raw conversation turns on write (so they can be vector-searched), and updates both the SDK and Function App plumbing to support searching either the “memories” container or the “turns” container. It also updates Cosmos container provisioning so the turns container is always created with vector + full-text policies (primed for future turn-search) and standardizes vector index type to quantizedFlat.

Changes:

Add ENABLE_TURN_EMBEDDINGS / enable_turn_embeddings toggle to control whether turn docs get embeddings on write.
Add target parameter to search APIs to route vector search to either memories or turns containers.
Update Cosmos provisioning (SDK + IaC) and tests/docs to reflect always-on turns vector/full-text policies and quantizedFlat vector indexes.

Reviewed changes

Copilot reviewed 27 out of 27 changed files in this pull request and generated 1 comment.

Show a summary per file

File	Description
tests/unit/test_thresholds.py	Adds unit coverage for `ENABLE_TURN_EMBEDDINGS` env parsing/defaults.
tests/unit/test_cosmos_memory_client.py	Updates provisioning expectations: turns container now includes vector/full-text policies and `quantizedFlat`.
tests/unit/test_container_routing.py	Adds tests for resolving `search(target=...)` to the correct container.
tests/unit/store/test_memory_store.py	Adds coverage ensuring turn embeddings are skipped by default and enabled when configured; validates search routing to turns.
tests/unit/function_app/test_pipeline_factory.py	Ensures FA pipeline wires `enable_turn_embeddings` into `MemoryStore`.
tests/unit/aio/test_cosmos_memory_client.py	Async equivalent of turns container provisioning assertions.
tests/unit/aio/store/test_memory_store.py	Async equivalent coverage for turn embedding toggle and search routing.
infra/modules/functions.bicep	Adds `enableTurnEmbeddings` parameter and app setting `ENABLE_TURN_EMBEDDINGS`.
infra/modules/cosmos.bicep	Switches vector index type to `quantizedFlat`; primes turns container with vector/full-text policies and updates excluded path pattern for turns embedding.
infra/main.parameters.json	Adds parameter wiring for `enableTurnEmbeddings`.
infra/main.bicep	Adds top-level `enableTurnEmbeddings` parameter and wires into functions module; updates embedding-dimensions description for `quantizedFlat`.
function_app/shared/pipeline_factory.py	Passes `enable_turn_embeddings` from config into `MemoryStore`.
function_app/shared/config.py	Adds `get_enable_turn_embeddings()` to read `ENABLE_TURN_EMBEDDINGS`.
function_app/local.settings.json.template	Documents and adds local `ENABLE_TURN_EMBEDDINGS` setting.
Docs/public_api.md	Documents new constructor arg `enable_turn_embeddings` and `search_cosmos(..., target=...)`.
Docs/concepts.md	Documents turn-embedding default/flag and how to search turns.
azure/cosmos/agent_memory/thresholds.py	Adds defaults + env parsing for enabling turn embeddings.
azure/cosmos/agent_memory/store/memory_store.py	Implements `enable_turn_embeddings` behavior + `search(..., target=...)` routing.
azure/cosmos/agent_memory/processors/inprocess.py	Wires SDK in-process processor store creation to the env-based toggle.
azure/cosmos/agent_memory/cosmos_memory_client.py	Adds `enable_turn_embeddings` constructor arg; primes turns container with vector/full-text policies; adds `search_cosmos(..., target=...)`.
azure/cosmos/agent_memory/aio/store/memory_store.py	Async equivalent: turn embedding toggle + `search(..., target=...)`.
azure/cosmos/agent_memory/aio/processors/inprocess.py	Async equivalent wiring for env-based toggle.
azure/cosmos/agent_memory/aio/cosmos_memory_client.py	Async equivalent constructor/provisioning/search changes.
azure/cosmos/agent_memory/_utils.py	Updates container policy builder: `quantizedFlat`, salience composite toggle, and excluded embedding path pattern.
azure/cosmos/agent_memory/_container_routing.py	Adds `resolve_search_target()` mapping public targets to containers.
azure/cosmos/agent_memory/_base/base_client.py	Adds base config wiring so `enable_turn_embeddings` comes from arg or env default.
.env.template	Documents and adds `ENABLE_TURN_EMBEDDINGS` default.

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

    indexing_policy = {
        "includedPaths": [{"path": "/*"}],
        "excludedPaths": [
            {"path": "/embedding/*"},
            {"path": "/source_memory_ids/*"},
            {"path": "/supersedes_ids/*"},
        ],


Copilot

Pull request overview

Copilot reviewed 25 out of 25 changed files in this pull request and generated 3 comments.

Comments suppressed due to low confidence (2)

infra/modules/cosmos.bicep:111

The containers’ indexingPolicy.excludedPaths no longer excludes the /embedding field. Embedding vectors are large arrays; indexing them in the standard index can significantly increase RU/index size without improving vector search (vectorIndexes/vectorEmbeddingPolicy cover vector search).

Recommend re-adding an excludedPaths entry for /embedding to keep indexing costs predictable.

        excludedPaths: [
          {
            path: '/source_memory_ids/*'
          }
          {

infra/modules/cosmos.bicep:196

Same as above for the turns container: indexing the /embedding array in the standard index can inflate RU/index/storage costs. Excluding /embedding from scalar indexing still allows vector search via vectorIndexes/vectorEmbeddingPolicy.

        excludedPaths: [
          {
            path: '/source_memory_ids/*'
          }
          {

aayush3011

Overall the PR looks, some changes to be made before it can be merged.

aayush3011 · 2026-06-25T21:37:45Z

        if not include_superseded:
            qb.add_is_null_or_undefined("c.superseded_by")

        query = f"SELECT * FROM c{qb.build_where()} ORDER BY c.created_at DESC"


We need to project explicit columns here excluding embedding. With enable_turn_embedding=True, get_thread() will return the vectors as well.

aayush3011 · 2026-06-25T21:38:36Z

        if not include_superseded:
            qb.add_is_null_or_undefined("c.superseded_by")

        query = f"SELECT * FROM c{qb.build_where()} ORDER BY c.created_at DESC"


We need to project explicit columns here excluding embedding. With enable_turn_embedding=True, get_thread() will return the vectors as well.

aayush3011 · 2026-06-25T21:46:01Z

+            store = MemoryStore(
+                containers=containers,
+                embeddings_client=embeddings_client,
+                enable_turn_embeddings=get_enable_turn_embeddings(),


Turn embeddings only ever get created in one spot: when the client saves a turn (through store.add() / store.push() ). That's the only code that looks at this switch.

Setting the enable_turn_embeddings doesn't do anything here, the value is picked from the client.

we should remove this.

aayush3011 · 2026-06-25T21:46:10Z

+            store = AsyncMemoryStore(
+                containers=containers,
+                embeddings_client=embeddings_client,
+                enable_turn_embeddings=get_enable_turn_embeddings(),


Turn embeddings only ever get created in one spot: when the client saves a turn (through store.add() / store.push() ). That's the only code that looks at this switch.

Setting the enable_turn_embeddings doesn't do anything here, the value is picked from the client.

we should remove this.

aayush3011 · 2026-06-25T21:46:29Z

+    store = MemoryStore(
+        containers=containers,
+        embeddings_client=embeddings,
+        enable_turn_embeddings=config.get_enable_turn_embeddings(),


Same as others

Turn embeddings only ever get created in one spot: when the client saves a turn (through store.add() / store.push() ). That's the only code that looks at this switch.

Setting the enable_turn_embeddings doesn't do anything here, the value is picked from the client.

we should remove this.

aayush3011 · 2026-06-25T21:46:57Z

    )


+def get_enable_turn_embeddings() -> bool:


Same as others

Turn embeddings only ever get created in one spot: when the client saves a turn (through store.add() / store.push() ). That's the only code that looks at this switch.

Setting the enable_turn_embeddings doesn't do anything here, the value is picked from the client.

we should remove this.

aayush3011 · 2026-06-25T21:47:36Z

          name: 'MEMORY_PROCESSOR_OWNER'
          value: memoryProcessorOwner
        }
+        {


Same as others

Turn embeddings only ever get created in one spot: when the client saves a turn (through store.add() / store.push() ). That's the only code that looks at this switch.

Setting the enable_turn_embeddings doesn't do anything here, the value is picked from the client.

we should remove this.

aayush3011 · 2026-06-25T21:47:47Z

 ])
 param memoryProcessorOwner string

+@description('Embed raw conversation turns on write so they can be vector-searched. The turns container is always provisioned with a vector index; this only toggles embedding generation. Default false.')


Same as others

Turn embeddings only ever get created in one spot: when the client saves a turn (through store.add() / store.push() ). That's the only code that looks at this switch.

Setting the enable_turn_embeddings doesn't do anything here, the value is picked from the client.

we should remove this.

enabled embeddings of raw turns, small other edits

8f4d858

Copilot AI review requested due to automatic review settings June 22, 2026 19:56

Copilot started reviewing on behalf of jcodella June 22, 2026 19:58 View session

Copilot AI reviewed Jun 22, 2026

View reviewed changes

Comment thread azure/cosmos/agent_memory/_utils.py

Comment on lines 391 to 397

indexing_policy = {

"includedPaths": [{"path": "/*"}],

"excludedPaths": [

{"path": "/embedding/*"},

{"path": "/source_memory_ids/*"},

{"path": "/supersedes_ids/*"},

],

jcodella added 3 commits June 22, 2026 16:07

fixed linting/gh review issues

8748072

small fixes

54279d2

added search_turns() for semantic search over raw memories

2aa2ff8

aayush3011 requested a review from Copilot June 25, 2026 15:16

Copilot started reviewing on behalf of aayush3011 June 25, 2026 15:17 View session

Copilot AI reviewed Jun 25, 2026

View reviewed changes

Comment thread azure/cosmos/agent_memory/store/memory_store.py

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

Comment thread azure/cosmos/agent_memory/_utils.py

Merge branch 'main' into main

954ff4e

aayush3011 requested changes Jun 25, 2026

View reviewed changes

Uh oh!

Conversation

jcodella commented Jun 22, 2026

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

Uh oh!

Uh oh!

Uh oh!

aayush3011 left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants