Describe the bug
In API v2, Episodic Memory episodes support metadata-based filtering and session-level isolation, but Episodic Summary does not.
As a result, all conversations under the same project share a single summary, causing cross-session context contamination.
Steps to reproduce
1.Create a project under a single org_id.
2.Start Conversation A under this project and send several messages to trigger episodic summary generation.
3.Start Conversation B under the same project and send different messages.
4.Retrieve the episodic summary for both conversations.
Result:
Both conversations return the same episodic summary, even though their episodes are isolated.
Expected:
Each conversation should have its own episodic summary.
Expected behavior
Episodic Summary should be isolated at the same granularity as episodes (e.g. by session / conversation / metadata).
At minimum, summaries should not be implicitly shared across unrelated conversations within the same project.
Environment
MemMachine version: v2 (current main branch)
API version: v2
Deployment: Self-hosted
Backend: PostgreSQL (short-term memory) + Neo4j (graph memory)
Operating System: Linux (Ubuntu 22.04)
Python version: 3.10
LLM provider: OpenAI-compatible API
Embedding model: OpenAI-compatible embedding model
Additional context
This appears to be a design inconsistency rather than a usage error.
Episodes and summaries expose different isolation semantics under the same API version, which is unexpected and error-prone.
Describe the bug
In API v2, Episodic Memory episodes support metadata-based filtering and session-level isolation, but Episodic Summary does not.
As a result, all conversations under the same project share a single summary, causing cross-session context contamination.
Steps to reproduce
1.Create a project under a single org_id.
2.Start Conversation A under this project and send several messages to trigger episodic summary generation.
3.Start Conversation B under the same project and send different messages.
4.Retrieve the episodic summary for both conversations.
Result:
Both conversations return the same episodic summary, even though their episodes are isolated.
Expected:
Each conversation should have its own episodic summary.
Expected behavior
Episodic Summary should be isolated at the same granularity as episodes (e.g. by session / conversation / metadata).
At minimum, summaries should not be implicitly shared across unrelated conversations within the same project.
Environment
MemMachine version: v2 (current main branch)
API version: v2
Deployment: Self-hosted
Backend: PostgreSQL (short-term memory) + Neo4j (graph memory)
Operating System: Linux (Ubuntu 22.04)
Python version: 3.10
LLM provider: OpenAI-compatible API
Embedding model: OpenAI-compatible embedding model
Additional context
This appears to be a design inconsistency rather than a usage error.
Episodes and summaries expose different isolation semantics under the same API version, which is unexpected and error-prone.