OpenTSLM time-series models land in SageMaker JumpStart
AWS adds Aionic Labs' open source time-series language models to SageMaker JumpStart, with a path to deploy and fine-tune them.
OpenTSLM lands in SageMaker JumpStart
On 17 November 2026, AWS announced OpenTSLM on SageMaker JumpStart, bringing Aionic Labs' open source family of time-series language models to its model hub. The models read operational measurements, from equipment sensors to supply and demand figures, alongside written context such as maintenance records and engineer notes. The post on the AWS Machine Learning Blog walks through deployment and a customer example.
What the models do differently
General purpose models such as Claude or ChatGPT can help analyze data, but the source notes that feeding long histories across hundreds of operational metrics as text or images can consume large numbers of tokens, which increases cost. OpenTSLM instead receives numerical time series next to text context and interprets how signals change across an observation window.
Nexthink's personal IT agent, Spark, uses this. It combines device telemetry such as CPU activity, memory use, available disk space, application stability and connectivity with timestamped events and device and software context. The model returns three outputs: a state summary explaining the likely issue and the signal patterns behind it, a recommended action listing checks to confirm the problem and a practical fix, and an expected outcome describing what should improve and how Spark will verify it.
In the post's low disk example, declining free space, a rising disk queue and more frequent application hangs line up with symptoms like a slow machine or trouble saving files. Spark checks available space, proposes an IT-approved cleanup, asks the employee to confirm, runs it, then checks whether the problem improved.
Deploying and adapting it
SageMaker JumpStart is a machine learning model hub where you can evaluate, compare and deploy pretrained foundation models. OpenTSLM can be deployed to SageMaker AI inference endpoints inside your own AWS environment. The source lists prebuilt inference containers with GPU acceleration, automatic scaling endpoints that follow request volume and latency needs, and a real-time inference API. To adapt the model to your own data, the post describes fine-tuning with TimeNet and a SageMaker training job. TimeNet supplies one shared format and API for locating, assembling and loading the datasets behind training and evaluation.
AWS positions JumpStart as complementary to Amazon Bedrock, which offers serverless API inference plus customizations, knowledge bases, agents and guardrails.
What it changes for builders
If an app already collects time-series data, this offers a pattern for turning that data into explanations and suggested next actions rather than dashboards alone. But it is an AWS deployment path: you need a SageMaker endpoint, a data pipeline and, for custom behavior, a fine-tuning job. Nothing in the source suggests a drop-in API for a small web app.
Limits and open questions
The announcement does not give model sizes, license terms, pricing, regional availability or benchmark results. It also does not quantify how many tokens or how much cost the time-series approach saves compared with text-based analysis. The Nexthink workflow is presented as an illustrative example, so independent performance data is not part of the post. How much data and compute fine-tuning requires is also unstated.
What it means for AI app builders
For people assembling apps with AI, the useful distinction is between models that reason over text and models that reason over measurements. If a product's value depends on sensor or telemetry history, a specialized time-series model may fit better than pasting numbers into a chat prompt. Tools like Draftship focus on the app layer, and its documentation on adding AI to your app covers connecting models to app logic. Draftship is not affiliated with AWS or Aionic Labs, and this announcement does not change what Draftship supports.
Primary sources
- Reason over time-series data with OpenTSLM on Amazon SageMaker JumpStart · AWS Machine Learning Blog, Nov 17, 2026
This article was drafted with AI assistance from the primary sources listed below and checked automatically against them. An editor hasn't reviewed it yet.