Skip to content

Self-evolving Context Database for AI Agents. Unify Agent Memory, Knowledge RAG and Skills.

fix(memory): truncate memory abstracts before vector writes (#2774)

chenjwe0ce25649a81
diff --git a/openviking/session/memory/memory_updater.py b/openviking/session/memory/memory_updater.py
index f4ede2412..65b74549f 100644
--- a/openviking/session/memory/memory_updater.py
+++ b/openviking/session/memory/memory_updater.py
@@ -44,6 +44,7 @@ from openviking_cli.utils import VikingURI, get_logger
logger = get_logger(__name__)
+_MEMORY_ABSTRACT_MAX_BYTES = 50_000
_EXTRACTION_CHUNK_MIN_CHARS = 100
_EXTRACTION_CHUNK_BOUNDARY_RE = re.compile(r"(\n+|[。!?;!?;]+|(?<!\d)\.(?!\d))")
_RESOURCE_ADDITION_FIELD_RE = re.compile(
@@ -1070,6 +1071,7 @@ class MemoryUpdater:
from openviking.session.memory.utils.link_renderer import LinkRenderer
abstract = LinkRenderer.strip_all_links(mf.content or "")
+ abstract = self._truncate_memory_abstract(abstract)
embedding_text = abstract
memory_type = uri_memory_type_map.get(uri)
@@ -1149,6 +1151,14 @@ class MemoryUpdater:
tracer.error(f"Failed to vectorize memory {uri}: {e}")
return attempted_count
+ @staticmethod
+ def _truncate_memory_abstract(abstract: str) -> str:
+ """Cap memory vector-store abstract fields below backend byte limits."""
+ encoded = (abstract or "").encode("utf-8")
+ if len(encoded) <= _MEMORY_ABSTRACT_MAX_BYTES:
+ return abstract or ""
+ return encoded[:_MEMORY_ABSTRACT_MAX_BYTES].decode("utf-8", errors="ignore")
+
async def generate_overview(
self,
memory_type: str,
diff --git a/tests/unit/session/memory/test_embedding_template.py b/tests/unit/session/memory/test_embedding_template.py
index cbf8de589..1912ef80f 100644
--- a/tests/unit/session/memory/test_embedding_template.py
+++ b/tests/unit/session/memory/test_embedding_template.py
@@ -366,3 +366,52 @@ class TestEmbeddingTextConstruction:
vector_text = mock_from_context.call_args[0][0].get_vectorization_text()
assert vector_text == "Trip summary"
+
+ @pytest.mark.asyncio
+ async def test_memory_abstract_truncated_to_50000_bytes_before_vector_write(self):
+ registry = MemoryTypeRegistry(load_schemas=False)
+ registry._types["events"] = registry._parse_memory_type(
+ {
+ "memory_type": "events",
+ "directory": "viking://user/{{ user_space }}/memories/events",
+ "filename_template": "{{ event_name }}.md",
+ "fields": [
+ {"name": "event_name", "type": "string"},
+ {"name": "content", "type": "string"},
+ ],
+ }
+ )
+
+ long_content = "你" * 20_000 # 60,000 UTF-8 bytes
+ updater = MemoryUpdater(registry=registry, vikingdb=Mock())
+ updater._viking_fs = Mock()
+ updater._viking_fs.read_file = AsyncMock(
+ return_value=MemoryFileUtils.write(
+ MemoryFile(
+ uri="viking://user/alice/memories/events/trip.md",
+ memory_type="events",
+ content=long_content,
+ extra_fields={"event_name": "trip"},
+ )
+ )
+ )
+ updater._vikingdb.enqueue_embedding_msg = AsyncMock(return_value=True)
+
+ result = MemoryUpdateResult()
+ result.add_written("viking://user/alice/memories/events/trip.md")
+ ctx = SimpleNamespace(user=None, account_id="default")
+
+ with patch.object(EmbeddingMsgConverter, "from_context") as mock_from_context:
+ mock_from_context.side_effect = lambda context: SimpleNamespace(
+ telemetry_id=None, id="msg-1", message=context.get_vectorization_text()
+ )
+ await updater._vectorize_memories(
+ result,
+ ctx,
+ extract_context=None,
+ uri_memory_type_map={"viking://user/alice/memories/events/trip.md": "events"},
+ )
+
+ memory_context = mock_from_context.call_args[0][0]
+ assert len(memory_context.abstract.encode("utf-8")) <= 50_000
+ assert memory_context.abstract.encode("utf-8").decode("utf-8") == memory_context.abstract