Smart Node 1.13.0 through 1.13.4 shipped between 1 and 8 November 2026. The main change: when an AI provider is unavailable, your agents now retry and route around the problem rather than leaving messages permanently sitting as unhandled. A role library with 100 pre-built roles, a faster Work Map, a browser-based installer, and fixes to chat reliability round out the week.
When the AI provider is unavailable
Before 1.13.0, a provider outage marked incoming messages as unhandled for good, and the agent was marked broken.
From 1.13.0, the messages are given back and the agent tries again: first after 60 seconds, then with each wait doubling, up to five minutes. The agent stays available the whole time.
A second change works alongside this: if the Claude account hits its limit, the node sends its Claude requests through OpenRouter for a cooldown period, so the rest of the crew keeps going. This only activates when a fallback model has been configured; it does not turn on by itself.
Together, these two changes mean the crew has a better chance of staying in motion through the kind of provider problems that used to bring everything to a halt.
A role library for setting up your crew
Before 1.13.4, the setup agent wrote every role from scratch each time. As of 1.13.4 it draws from a library of 100 pre-built roles across 18 departments. It reads each relevant role in full before making proposals, and explains what each agent would do and what its first task would be, in the owner's own words.
A new Personal department adds roles including Personal Assistant, Legal Guide, Health Guide, Tech Helper, and Wellbeing Companion. Each of these carries a clear disclaimer that the role is not a substitute for a licensed professional.
Other changes in this week
The Work Map now loads in roughly 1.5 seconds. Before 1.13.1 it took about 37 seconds. It reads far fewer records and reads the rest in parallel.
A browser-based installer arrived in 1.13.3. It opens within a second, shows each step with live output as the install runs, and can be stopped mid-run. One limitation carried through every release in this series: the module-selection step did not pass the chosen configuration to the actual installer. Every install in 1.13.x installed all modules regardless of what was selected. This is fixed in 1.14.0.
Reminders are now part of the core and no longer require the Tasks module to be installed.
Chat and sign-in fixes. The chat reconnects when you return to a stale tab, so "Thinking" and "Waiting for an agent to start" indicators no longer persist after an answer has already arrived. Sending the same message twice now returns the original reply rather than creating a second inbound message. Sign-out from Studio and the panel works again, having previously been blocked by a missing token. Password change is reachable from Studio's Settings page, and the "Forgot password?" option now appears on local nodes.
On first run, the owner's password is shown once with a copy button, Before, the node created a password nobody could ever see.
After an update, agents no longer lose write access to their own folders. An install step had taken the agents' own working folders away from them, so after every update an agent could not save its work. This is fixed in 1.13.1.
Studio replaced "Workspace" as the product name throughout.
Known issues at this release
Three issues were open at 1.13.0. Agents in a chain do not stop autonomously and may invent decisions. The "Waiting to start" label in the chat queue can persist for up to 24 hours if an agent cannot start. A user with the standard user role can view and cancel reminders that belong to other users.
Get this release
The full list of changes is in the changelog.