Nightly learning round
Every night the bot reads your conversations from the day, records what it learned and turns repeated things into fixed rules.
Today
Learning and training
A system that gets better with you.
Every Node.Bot learns. It makes the odd mistake, like any new colleague. Explain what was wrong and it adds that to its rules overnight. What repeats gets faster over time, because it works out its own routines for it.
The learning loop
Work by day. Reflect by night.
Work
The bot does its tasks, answers questions and remembers what went well and what did not.
Explain
If something was wrong, tell it in plain words: “The hallway light means the ceiling lamp, not the wall light.”
Adjust overnight
During the night the bot goes through its conversations and writes rules from them, which it follows from then on.
Get faster
What often runs the same way becomes a fixed command or a new skill, which you approve.
What already learns today
Five ways to learn.
Learned commands
A sentence that was run or confirmed the same way three times will in future run without a language model and without waiting.
Today
New skills
If the bot spots recurring routines, it suggests a new skill. It only becomes active once you agree.
Today
Getting to know the home
The home bot learns which sensors belong to which room and topic, and writes notes about the household overnight.
Today
Its own personality
A bot's tone and manner sharpen with its conversations. The core stays, the fine details adapt.
Today
Faster over time
Many steps become one.
On day one, the bot thinks afresh about every request. Later it knows the way. Pick a point in time and see how a recurring routine changes.
Schematic, without time figures. How fast it gets depends on the task.
Example: “Hallway light off”
- Understand the sentenceThe language model works out what is meant.
- Follow-up question“Ceiling lamp or wall light?”
- Find the deviceLook for the hallway light in the home.
- SwitchAfter your approval.
- Fixed commandRuns without a language model, instantly.
On the first day, the bot asks which lamp is meant.
You explained that the ceiling lamp is meant. Overnight this became a rule, so the follow-up question is dropped.
The sentence was often run the same way and confirmed. Now it is a fixed command and runs without a language model.
Training for businesses
Onboard it like a new employee.
A bot that talks to your customers is trained first. You give it access to catalogues, manuals and procedures. From these it builds a second brain: a knowledge network in which every statement has its source. Then you practise with it how it should deal with customers.
- Hand over documentsCatalogues, price lists, manuals, frequent questions.
- Build the knowledge networkA staff member confirms product facts before the bot states them.
- Practise and checkTrial conversations with the team, then approval.
- Keep learning on the jobAnything it does not know for sure, it passes to a person.
Planned Onboarding comes after version 1.0. Knowledge networks for devices and sources run today in observe mode.
You stay in control
Learning with limits.
Visible Today
What the bot has learned is in its rules. You can read it.
With approval Today
New skills and new fixed commands only become active after you confirm.
Local Today
Learning uses the model on the Node. A cloud model only if you explicitly allow it.
Report mistakes precisely In progress
Mark a mistake, get the cause and a remedy, and approve the remedy.
Write its own tools Idea
For demanding, recurring tasks, the bot builds checked tools for itself.
Only from authorised people Today
Today the bot only learns from the owner's conversations. Learning from the feedback of other authorised people is planned.
A colleague who learns something new every day.
Sign up and tell us what your bot should learn first.