AtMem 2.3.5

Stable release. AtMem 2.3.5 simplifies local onboarding and makes the AtMem and AtFlows dashboards easier to navigate together. atmem init now starts the AtMem dashboard, the pinned AtFlows companion in AtMem-delegated login mode, and AtBot where available. atmem status reports installed and running versions, actual local URLs and ports, and actionable startup errors. If the preferred dashboard port is occupied, initialization chooses and reports an available loopback port. The AtMem dashboard links to the detected AtFlows dashboard; AtFlows 0.1.2 links back when delegated login is configured. Browser titles and favicons use the AtMem.ai product identity. AtFlows server logs now display local time with an explicit offset.

Install and upgrade

Back up or snapshot a quiescent AtMem Home before upgrading. In the Python environment used by AtMem:

python -m pip install --upgrade atmem==2.3.5
atmem init
atmem status

For an existing OpenClaw connection, also run:

atmem openclaw upgrade
atmem control verify

For a new OpenClaw connection, use atmem openclaw install after initialization. AtMem 2.3.5 pairs with openclaw-memory-atmem@2.3.5, pins atflows==0.1.2 and keeps atmem-atbot==0.1.0. Python 3.10–3.13 is supported. Running AtFlows needs Bun 1.1 or newer and may need network access on its first launch to prepare its bundled runtime. Installing the wheel alone does not start services or collect telemetry.

atmem init is safe to repeat: it preserves existing AtMem Home, memory, evidence, AtFlows data, and account credentials. It does not reissue an existing Administrator password. On upgrade, it restarts a stale AtMem dashboard and checks each companion. If AtFlows or AtBot cannot start, it reports the error; AtMem's core memory remains available. Run atmem status to see the current state and URLs rather than assuming the default ports. A pre-existing independently started AtFlows instance is reported, not silently taken over. Review any existing standalone AtFlows account and authentication setup before switching to the AtMem-managed instance.

What changed from 2.3.4

• One initialization command starts and checks the local dashboards and AtBot companion. Status shows version and port mismatches with a repair path.
• The AtMem dashboard exposes a link to a healthy local AtFlows dashboard; the delegated AtFlows dashboard links back to AtMem. Both show consistent browser identity.
• AtFlows 0.1.2 logs server timestamps in local time with an explicit offset.

Boundaries and limitations

AtFlows remains a separate local observability service and database. Starting it does not configure a client to send telemetry, authenticate its OTLP receiver or model proxy, convert traces into canonical memory, or enable the read-only review-lead handoff. The latter remains an explicit investigation using atmem atflows review and exact session evidence. AtMem does not alter OpenClaw or another host unless you run that adapter's installation or upgrade command. Memory injection remains governed by the existing shadow/active state; initialization does not activate it. Keep local listeners on trusted loopback interfaces and review any reported startup error before relying on the companion dashboards.