Open source OpenTSLM model family for time-series analysis available on Amazon SageMaker JumpStart
The company Aionic Labs has released OpenTSLM, an open source family of models for time-series analysis, available on Amazon SageMaker JumpStart. The OpenTSLM-Flamingo model (5.24 billion parameters) combines Gemma 3 4B with a time-series encoder; Nexthink uses it in its IT agent Spark.
The company Aionic Labs developed OpenTSLM, an open source family of language models for time-series analysis (time-series language models, TSLM), which is now available on Amazon SageMaker JumpStart for deployment in your own AWS environment. The models are designed to interpret operational measurements – from device sensor data to supply and demand – together with contextual information such as maintenance records, market reports, or technician notes. The models use the shared TimeNet infrastructure, which, according to the company, provides a unified format and API for searching, creating, and loading datasets for training and evaluation.
The specific model available in SageMaker JumpStart is called OpenTSLM-Flamingo. It combines the Gemma 3 4B language model with a time-series encoder and cross-attention layers, which allow the language model to work with the encoded data series. The model has a total of 5.24 billion parameters stored in bfloat16 format, which, according to the source, corresponds to approximately 10.5 GB of GPU memory.
As a concrete use case, the article mentions the collaboration between Nexthink and Aionic Labs on the Spark tool, Nexthink's personal IT agent. Spark combines device telemetry (CPU load, memory, free disk space, application stability, connectivity) with time-stamped events and device and software context. According to the company, the model can recognize a problem based on a combination of these signals, suggest a specific solution to the employee (for example, clearing disk space), and after it is carried out, verify whether the condition has improved, or otherwise hand the case over to IT support.
As motivation, the source states that general-purpose language models such as Claude or ChatGPT can analyze data, but processing long histories across hundreds of operational metrics in the form of text or images consumes a large number of tokens and increases costs. TSLM, by contrast, accepts numerical measurements directly together with textual context.
Why it matters
For developers working with operational data (sensors, telemetry, demand), OpenTSLM offers a pretrained model deployable in your own AWS environment that processes numerical time series directly – without converting them to text, which would increase token consumption and costs with general-purpose language models. According to the company, this makes it possible to build applications for proactive diagnostics, as shown by its deployment in Nexthink's IT agent Spark, where the model combines device telemetry with context and suggests a specific fix before the user even has a chance to complain.
Two audiences, two different impacts
What this means
For individuals
For developers and data specialists, this means an accessible pretrained model that processes numerical time series directly together with textual context, without the need to convert the data into text for a general-purpose language model and pay for a high number of tokens.
For a business
Companies with operational data (IT support, manufacturing, energy) gain an accessible model for proactive diagnosis of problems before they are reported – as shown by its deployment in the Spark tool from Nexthink, which combines device telemetry with context and suggests a specific intervention, thereby, according to the company, reducing the need for IT support intervention.
DevelopmentCheck the original
Event sources
clearly official source · 1 publisher, 0 independent. We count feeds from the same owner only once.