Datastreams are the new standard for bringing external datasets into the Tibo platform.
They replace the older Datasource concept and add automatic AI training on your data, so you get smarter simulations with very little setup.
Use this article to learn what datastreams are, how to upload data, how AI training works, and how to use a datastream in your simulations (including simple “baseline” scenarios).
What is a Datastream?
A datastream = your data + an AI model trained on that data.
Unlike a classic dataset upload, every datastream you create is automatically analyzed and used to train a custom AI model. This model is tailored to your data and learns patterns such as:
Seasonality (daily, weekly, yearly effects)
Calendar effects (holidays, special days)
Temperature sensitivity and other external influences
Once the model is trained, you can use the datastream directly in your simulations.
Datastream Lifecycle & Statuses
After you upload data, the datastream moves through several internal states. In the interface, you may see messages like:
Processing csv
Your CSV file or pasted data is being validated and converted into a time series.Datasource is ready
The raw data is available in the system and can already be used as a basic input.Model Training
The AI model is being trained on your datastream using the chosen parameters.Model Ready
The AI model is fully trained and ready to be used in simulations.
You can continue working while training is in progress.
Model training typically only takes a few minutes and runs in the background.
Uploading Data into a Datastream
You can create or update a datastream using either of these methods:
1. Drag and Drop a CSV File
Go to the Datastreams section in Tibo.
Click New datastream (or the relevant “Add” button).
Drag and drop your CSV file into the upload area, or select it from your computer.
Confirm the detected columns and timestamp format if prompted.
Continue to Advanced settings (optional) and then save.
Requirements (typical setup):
Data should represent a historical time series (e.g. hourly, daily, monthly values).
Each row usually contains a timestamp and one or more numeric values.
2. Paste Data Manually
Go to the Datastreams section.
Start a new datastream.
Switch to the Paste data input option.
Paste your time series data into the text area (e.g. copied from Excel).
When prompted, specify the time range (e.g.
01-01-2024to31-12-2024).
This helps Tibo align the pasted values to the correct timestamps.Confirm the structure, adjust any settings if needed, and save.
Both upload methods trigger the same AI training process in the background.
Advanced AI Training Settings
Before you save the datastream, you can optionally configure advanced training parameters.
Default Recommended Settings
By default, Tibo pre-selects four recommended training parameters.
These defaults are suitable for most use cases and are designed to give robust results without extra tuning.
You can enable additional options if you want more control or have a specific use case.
Smart Parameter Handling
Even if you enable extra advanced options, Tibo may decide not to use some of them in practice. For example:
If the data period is too short, certain seasonal or long-term effects can’t be reliably learned.
If enabling a setting would lead to overfitting or unrealistic extrapolation, it may be ignored.
This automatic handling ensures that:
The trained model remains stable and reliable.
You don’t need deep machine learning expertise to get good results.
When Does Training Start?
Training starts automatically once data is uploaded and the datastream is saved.
You don’t need to manually trigger it.
While it’s training, you can already continue configuring your digital twin or other parts of your project.
How the AI Model Behaves
The AI model behind a datastream is built to generalize structural behavior from your historical data. Here are some key capabilities:
Seasonal and Calendar-Based Effects
The model can recognize and reproduce patterns that recur over time, for example:
Weekday vs. weekend patterns
Monthly or annual cycles
Holidays and special days
It can also handle shifting dates. For instance:
If New Year’s Day falls on a Sunday in your training data, but on a Monday in a future year, the model still understands that this is a similar type of day and predicts accordingly.
Temperature Sensitivity
If your data is influenced by weather (e.g. energy consumption, cooling demand), the model can:
Detect correlations between temperature and your values.
Apply similar behavior when future temperature forecasts are fed into the simulation.
For example, if your data peaks around 30°C, the model will tend to simulate similar peaks when high temperatures are forecasted again.
Using a Datastream as a Baseline (Exact Year Reproduction)
Sometimes you don’t want predictions at all—you just want to replay a historical year exactly as it was. For example, to create a reference scenario or validate a model.
For that, you can use Exact Year Reproduction (Baseline Mode):
Create and train a datastream as usual.
When configuring your simulation, select:
The original year of your historical data as the simulation period.
Enable the option to use the datastream in baseline mode (exact year reproduction).
In this mode:
Tibo reproduces your uploaded datastream 1-to-1.
No additional AI predictions or extrapolations are applied.
The simulation shows the raw input values as they occurred historically.
This is especially useful for:
Reference scenarios (“What if conditions were exactly like 2023 again?”)
Validation and comparison against AI-generated scenarios.