Sift launches AI agent to tackle hardware telemetry

Sift has launched Sift Agents, an artificial intelligence agent for investigating hardware telemetry. The feature is in research preview with selected engineers at existing customers.

The product is designed to analyse test, flight and production data within a customer's Sift environment and save the results as reusable rules and reports. It draws on a programme's assets, runs, channels and existing rules to investigate anomalies and record findings for engineering teams to use later.

The launch targets a persistent problem in hardware engineering: test and production systems generate more sensor data than teams can review manually. Many investigations never reach a root cause because engineers would need weeks to rule out every possible failure model by hand.

The agent runs inside Sift's platform rather than sending data to an outside system. That is intended to support customers handling sensitive programmes, including export-controlled work on the company's GovCloud deployment.

It also inherits existing user permissions and requires human approval before making changes, placing the tool in a review role rather than giving it free rein over engineering data and workflows.

Manual bottleneck

Hardware developers in sectors such as aerospace, defence and robotics often run large-scale tests that produce telemetry streams from millions of concurrent sensors. While software teams have spent years building observability practices around system logs and performance data, hardware teams have generally relied on slower, more manual methods to inspect results and compare runs.

Sift is positioning the new agent as a way to speed up that work. Engineers can use it to test hypotheses, identify anomalies and turn findings into records that can be reused in later investigations.

“Hardware teams have more data than engineers, and every test widens the gap,” said Austin Spiegel, Co-Founder and Chief Executive Officer of Sift.

“They need an agent that already knows their machines, their channels, and their rules, and that works inside the platform their program already trusts,” Spiegel said.

Alongside Sift Agents, Sift is also adopting the Model Context Protocol, or MCP, which allows engineers to connect external artificial intelligence tools directly to Sift. The standard is model-agnostic, meaning users can work with a range of platforms including Claude Code, Cursor and Codex.

MCP is available to all customers at no additional cost. The MCP server runs on the engineer's own machine under that user's API key and permissions, giving teams another way to query and analyse telemetry without moving it out of their own working environment.

Customer example

Plantd, which makes carbon-negative building materials, has been using Sift as its production telemetry system. According to the company, Sift MCP reduced the time required for one post-test investigation from about 40 hours to an afternoon.

Previously, that work involved manually exporting channels and rebuilding statistics, graphs and written reviews. The faster turnaround meant engineers could explore questions that would once have taken too long to justify.

“When it took me half a day to get an answer, I only asked the questions that were worth half a day,” said Matthew Barr, Process Engineer at Plantd.

“Now if I want the median instead of the average, or I want to throw out the startup transient, it's one sentence. So I end up with better answers, not just faster ones,” Barr said.

Sift was founded by former SpaceX engineers and sells software to hardware teams that need to collect, organise and analyse telemetry across development and operations. Its customer base spans industries where machines, vehicles and industrial systems generate large data sets that are difficult to interpret quickly.

The launch of Sift Agents reflects a broader shift in those sectors as companies look for tools that can help engineers keep pace with rising data volumes. In hardware development, where a delayed diagnosis can slow testing schedules, production output or flight readiness, the value of automation often lies in narrowing the gap between data collection and a usable explanation.

The new agent saves findings back into Sift's platform as durable artefacts, allowing engineering teams to reuse rules and reports across future investigations.