Scheduling and Maintenance

Every long-running system accumulates recurring chores: pull the numbers, tidy the queues, retrain the models, clean the logs. Most stacks scatter those across cron entries that someone has to remember exist. Auto Learning Agents gives them a single supervised home, the schedule node, which means the chores are visible in the same activity feed as everything else, pausable like everything else, and impossible to forget about because they are part of the system rather than around it.

Hourly: Reports and Evidence

Every hour, the node pulls the measurement feeds you have connected: Google Analytics traffic, Search Console search performance, and social analytics across the connected platforms. Then it does the step that raw dashboards skip, cross-referencing the evidence: what was published, what was engaged with, what drove traffic, lining results up against the work that might explain them. This is the data layer behind honest answers, when you ask the master agent whether this month's content push is working, the response rests on hourly-fresh numbers already correlated with activity.

Every Six Hours: Area Maintenance

Each platform area, marketing, customer service, social media, and the rest, gets a maintenance pass on a six-hour cycle: queues groomed, states reconciled, the small housekeeping that keeps an area healthy between the bigger moments. Like a well-run shop's closing routine, it is invisible when it works, and what it surfaces, anything irregular, lands in the activity feed or becomes a flag.

Every Twelve Hours: ML Training

Twice a day, the node triggers training for the platform's ML layer, the topic classifier and anomaly models that the learning system maintains, refreshed against the latest accumulated data. The cadence keeps the models current with how your install actually talks and behaves, without training so often that the system churns.

Daily: Cleanup

Once a day, cleanup scripts from the cron/ directory run: logs rotated, stale artifacts cleared, temporary files retired. The directory is the extension point, drop a script into cron/ and it joins the daily run, which is the intended home for any recurring chore your install grows that does not need AI judgment, just reliable execution.

Weekly, Optionally: Fine-Tuning

With trainOllama enabled in settings, the node adds a weekly fine-tuning run that trains a local Ollama model on your install's own memory data, the accumulated knowledge, procedures, and outcomes your agents have gathered. Week over week, that produces a local model increasingly fluent in your domain, your products, and your way of working, the deepest form of the platform's learning loop, and one that never sends your data anywhere. The AI models guide covers where a fine-tuned local model fits in the lineup.

Three Kinds of Time

The platform has three timing mechanisms, and knowing which owns what makes the system easy to reason about. Agent ticks are your defined work: each agent wakes on its own schedule to do its job. Impulse is judgment on a timer: it reviews, notices, and initiates, the closest thing to a mind on a schedule. The schedule node is the mechanical layer: pulls, maintenance, training, and cleanup that need reliability rather than creativity, most of its work runs without an AI call at all. When you wonder where a new recurring need belongs, ask what it requires: a role and a model means a new agent, judgment about the whole system means Impulse's instructions, and pure clockwork means a script in cron/.

Running It

The schedule node is a supervised node like every other: it ticks hourly by default, appears in the activity feed when it works, honors the global pause, and restarts automatically if anything stops it. The cadences above are the defaults and are configurable. There is rarely anything to do here, which is the point, but the visibility matters: when you wonder whether the reports pulled or the cleanup ran, the answer is in the feed, timestamped, rather than in a server's crontab that no one has looked at since installation.