AtMem 2.3.13

Stable dashboard refresh, coordinated with AtFlows 0.1.6, AtSys 0.3.3 and the refreshed AtMem.ai website. Governed memory creation, scoped retrieval, context provision, encrypted evidence and the research opt-ins keep their existing behavior and defaults.

What changes

• The local AtMem dashboard (atmem ui) uses the AtMem.ai Architecture Canvas theme: self-hosted DM Sans and Manrope fonts (SIL OFL, served by the local dashboard only), softer floating surfaces, segmented navigation and the AtMem.ai mark with the "AtMem.ai | Agent Memory" name and page title.
• The dashboard opens in the light theme. Dark mode is used only when you choose it with the theme button; the choice is stored in the browser. Previously a first visit followed the operating system colour scheme.
• The pinned AtFlows companion is the stable atflows==0.1.6, which brings the same theme to the AtFlows dashboard without changing collection or login.

Install or upgrade

Python 3.10–3.13:

python -m pip install --upgrade atmem==2.3.13
atmem --version
atmem status

Optional coding memory:

python -m pip install --upgrade 'atmem[code]==2.3.13'
atmem code --help

Back up the AtMem home, database, encryption keys and evidence before upgrading. Let active tasks finish, stop services you own, upgrade, then restart connected clients and dashboards. Existing data is retained and no migration is needed. Installing does not call a model, activate a provider or enable research routing. Persisted upgrades from 2.3.12 and 2.3.11b1 are qualified in CI before tagging.

Exact compatibility

• atmem==2.3.13
• atmem-atbot==0.1.1 (unchanged legacy memory-proposal companion)
• atflows==0.1.6 (Architecture Canvas dashboard; behavior unchanged)
• openclaw-memory-atmem@2.3.13 (matching npm bridge; behavior unchanged)

Upgrade the OpenClaw integration through the Python-owned installer:

atmem openclaw upgrade
atmem openclaw status

The standalone atbots agent platform is a different distribution, not a replacement for atmem-atbot. AtSys is separately installed; native workspace supervision supports macOS/Linux. AtMem and AtFlows retain their supported Windows/Linux/macOS paths; native Hermes integration is unavailable on Windows.

Research results and boundaries

The final self-tested DolphinBench run passed 479/600 tasks (79.83%) on its documented frozen research artifact (routed-v4 source with the installed 2.3.10b1 base). It is not a fresh benchmark of the 2.3.13 wheel, and no new scored run is part of this release.

This release is a visual change to the local dashboards. It does not change memory authority, retrieval, evidence or policy. Retrieval is not proof an agent used a fact or learned. Website source review, owner merge and production deployment are separate publication states. 2.3.13 is the current stable installation target.