How does the bot get better overnight?

Learning that never surprises you

Bots that silently rewrite themselves are hard to trust. BotableX learns from every conversation, but the learning lands in two places that you control, and behavior changes are inert until a human approves them. You can see exactly what the bot proposed, why, and what happened to it.

What counts as a candidate

During the day, several things become candidates for learning: a teammate’s guidance during a consultation, a correction typed after a takeover, a customer confirming that an answer solved their problem, and a customer saying it did not. Human agents can also mark a conversation as Always learn or Don’t learn, and leave a short training note explaining what should be taken from it.

The nightly run

Once a night, or whenever you press Update learning now, BotableX processes the day’s candidates. Each is judged for usefulness, then classified along one axis:

• A fact. Something true about your business that the bot did not know or had wrong: an opening time, a returns window, a product limitation. Facts are added to the bot’s Learning Bank, a system-managed Knowledge Bank the bot reads like any other. You can switch the Learning Bank off for a bot at any time.
• A behavior. Something about how the bot should act: tone, when to offer a callback, how to handle a recurring situation. Behaviors are never applied directly. They become proposed notes for the bot’s Agent Core and appear under Suggestions awaiting your review.

Reviewing suggestions

Open Learning for the bot. Each suggestion shows the proposed note, the conversations that produced it, and how many times the pattern occurred. Press Approve to adopt it; the bot follows it from the next conversation. Press Discard to drop it. Discarded suggestions are remembered so the same pattern is not proposed again the next night.

Suggestions that contradict something you wrote yourself in the bot’s Agent Brain are listed separately under conflicts, with both texts side by side. Your hand-written rules always win until you decide otherwise.

The four tiers of what the bot knows

It helps to know where each kind of knowledge lives:

Tier	What it holds	Who changes it
Tenant Mind	Your organization’s identity, voice, prohibited topics and escalation rules, shared by every bot	You, in Global Mind
Agent Brain	A bot’s own instructions, limits and work methods	You, in Bots
Agent Core	Notes the learning run proposed and you approved	The nightly run proposes, you approve
Agent Memories	Short-term memory about an individual customer	The bot, automatically, with retention limits

Customer memories are kept for a limited time: 30 days for anonymous website visitors and 180 days for identified customers, after which they are deleted.

Teach from past chats

If you have months of history before learning was switched on, press Teach from past chats. The run reads the recorded conversations and produces candidates from them, following the same fact-or-behavior rules and the same approval gate.

Watching it work

Learning activity is the decision trail. It lists every candidate the run saw, what it was classified as, whether it became a fact or a suggestion, and whether you approved or discarded it. If the bot ever does something you did not expect, this is the first place to look.