OpenTSLM - A community for Time Series Language Models (TSLMs) and Time Series Foundation Models (TSFMs)
In October 2025, we released a preprint about a novel multimodal LLM architecture that can handle time series as a native modality: Time Series Language Models (TSLMs)
OpenTSLM: Time-Series Language Models for Reasoning over Multivariate Medical Text- and Time-Series Data
Since it's release, there have been people all around the globe building on our open-source models and implementations across academia, industry and hospitals. Peopke created models for automated ECG interpretation deployed inside of hospitals, monitoring of patients in the ICU, manufacturing and predictive maintenance applications, wearable data interpretation, and optimization of machines in the energy sector. This includes people across Stanford, Harvard, Dartmouth, NUS, ETH Zurich, KAIST, Seattle Children's Hospital, University of Zurich, as well as Google, Meta, Microsoft, and AWS.
Important
We want to bring together the community on TSLMs all around the world to push this field forward, to build foundational TSLMs that input and output text and time series: AI that understands temporal sensor data natively, incorporates context, turns signals into action, and enables prompt-based interaction with and generation of time series data. We are supporting any research on TSLMs.
Time-series data is the most common modality measuring the physical world. It captures heart rhythms, vital signs, wearable signals, industrial machinery, energy systems, environmental conditions, transportation networks, financial activity, and countless other processes that evolve over time.
Yet general-purpose AI systems do not understand time series natively. Temporal signals are often reduced to summary statistics, converted into images, or processed by separate specialized and rigid models before they can be used by a language model.
We believe time series should become a first-class modality for foundation models.
A Time Series Language Model should be able to:
- Combine time series with text and other modalities
- Interpret raw and multivariate temporal signals
- Explain patterns, anomalies, and predictions in natural language
- Answer questions about historical and real-time measurements
- Follow instructions grounded in sensor data and context
- Generate forecasts, simulations, and synthetic time series
- Transform and edit time series based on instructions
- Recommend actions based on temporal evidence
- Generalize across tasks, datasets, sensors, and domains
Our long-term vision is an open ecosystem of multimodal foundation models through which people can interact with temporal data as naturally as they interact with text.
Today, research, data and knowledge around TSLMs and TSFMs is fragmented. We build and maintain open foundations that help researchers develop, evaluate, and deploy such models.
With OpenTSLM, we provide open-source models, architectures, datasets, training code, evaluations, and reference implementations for building AI systems that understand time series as a native modality.
With TimeNet, we provide standardized data and training infrastructure — including scalable infrastructure for foundation model training, common data formats, a growing cross-domain corpus of datasets, a Python SDK, benchmarks, and evaluation tools for TSLMs.
We support open research across the entire field—from new architectures and multimodal reasoning to generation, forecasting, interpretability, safety, and applications in healthcare, wearables, manufacturing, energy, climate, finance, and beyond.
We are industry veterans from leading frontier AI companies in tech, and researchers from academia, with experience building and deploying large-scale AI systems.
We are bringing that experience together to scale OpenTSLM, grow the community, and build the shared infrastructure needed to move the field forward.
The wider OpenTSLM community includes contributors and collaborators across institutions such as Stanford, Harvard, Dartmouth, NUS, ETH Zurich, KAIST, Seattle Children’s Hospital, the University of Zurich, Google, DeepMind, Meta, Microsoft, and AWS.
We are an open community with no gatekeeping whatsoever. We share our infra, knowledge, and experiences. We have a frequent exchange, both physically and virtually, with TSLM researchers all around the world. If you are interested, please join us! Our credo is sharing is caring. If you are working on anything that you think is relevant or interesting for the community, please join and share, and we will support. We welcome everyone.