Sift Agents
Sift Agents investigate your hardware data, connect findings to the evidence, and help turn what you learn into reusable checks.
See how Orbital Ops cut analysis between tests from 30+ minutes to ~1 minute
Simulated thermal
Pressure transient Clip 00:05.0 to 00:06.2
Sift Agents in action
Move from an unexpected test result to evidence you can inspect, answers you can share, and checks you can reuse.
01 / Ask
Engineer: Was that pressure spike expected?
Run 12Interval 00:05.0 to 00:06.2
02 / Investigate
The spike matches the reference startup pattern. Pressure returns within range before startup ends.
Matches reference startup · in range after startupCompares chamber pressure with one reference run.
Video · Camera 1Open in Explore
Simulated thermalSame frame
SelectedSaved with review v1 00:05.0 to 00:06.2
03 / Act
Engineer: Create a check for sustained pressure deviations after startup.
The check excludes startup and flags a sustained pressure deviation in Run 6.
Review saved. Check saved as a draft.
Engine startup review · v1
Flag pressure that remains outside the expected range after startup.
Evaluated during mainstage92 to 104 bar · 0.5 s
00:0000:07.000:10
Run 6 · 00:08.1 to 00:09.0
Startup cleared.Run 6 deviation flagged.
| 10 demo runs | Hits | Window |
|---|---|---|
| Run 12 | 0 | Startup spike not evaluated |
| Run 9 | 0 | None |
| Run 6 | 1 | 00:08.1 to 00:09.0 |
| Runs 1 to 5, 7, 8 | 0 | None |
persistence(($1 < 92 || $1 > 104) && $2 == 'MAINSTAGE', 0.5)Evaluates only while the phase is MAINSTAGE, so startup, shutdown, and off periods are excluded. A hit needs pressure outside 92 to 104 bar for at least 0.5 s. Saving it as a draft does not enable it.
| Clip time | Observation | Bar |
|---|---|---|
| 00:00.0 | Phase STARTUP; rule not evaluated | 40.0 |
| 00:05.0 | Selected interval opens | 97.8 |
| 00:05.5 | Peak, 2.1 bar above the reference peak | 113.4 |
| 00:05.8 | Back under 104 bar | 102.3 |
| 00:06.2 | Selected interval closes | 96.1 |
| 00:07.0 | Phase MAINSTAGE; rule evaluates 92 to 104 bar | 97.9 |
| 00:10.0 | End of clip window; no sample outside the range | 97.1 to 98.5 |
04 / Reuse
Engineer: Check Run 13 against the saved review.
Engine startup review · v1
Run 13 matches the saved startup pattern. No sustained pressure deviations flagged.
Matches saved startup pattern · peak 112.0 bar at 00:05.6
Saved check: 0 hits · 97.2 to 98.7 bar after startup
Customer results
“When we’re running tests back to back, I can check a hunch against the data in minutes and use what I learn in the next test.”
“This is saving me dozens of hours a week, easily!”
Customer-reported results for the workflows shown.
Across the hardware lifecycle
Example Sift Agents conversation with demo data.
Engineer: What changed between these two builds?
Sift AgentsSifting…
Sift Agents: Build B’s motor controller runs 6 °C warmer at full load. Bus voltage and vibration match Build A within tolerance.
Controller temperature · full loadBuild A vs Build B
Bus voltage and vibration within tolerance
Engineer: Show me the full-load comparison for Build A and Build B.
Sift AgentsSifting…
Sift Agents: Here is the comparison on the same test profile. Controller temperature differs at full load; bus voltage and vibration match within tolerance.
Build A vs Build BSame test profile
Controller temperature · full load
Engineer: Draft a comparison review for the next design iteration.
Sift AgentsSifting…
Sift Agents: Drafted a review that keeps this comparison for the next design iteration. Nothing is saved until you approve it.
Build A vs Build B reviewDraft · not saved
Engineer: Which runs failed our acceptance criteria?
Sift AgentsSifting…
Sift Agents: 3 of 24 runs failed. Runs 7 and 15 exceeded the vibration limit, and run 19 fell short of the minimum thrust hold.
Acceptance criteria · 24 runs3 failed · 21 passed
Engineer: Show me which acceptance criteria each failed run missed.
Sift AgentsSifting…
Sift Agents: Each failed run, with the criterion it missed. The other 21 runs passed.
Failed runs · acceptance criteria3 of 24
| Run | Criterion | Result |
|---|---|---|
| Run 7 | Vibration limit | Exceeded |
| Run 15 | Vibration limit | Exceeded |
| Run 19 | Minimum thrust hold | Fell short |
The other 21 runs passed
Engineer: Draft a test review with these results and acceptance criteria.
Sift AgentsSifting…
Sift Agents: Drafted a test review that references these criteria and results, so the next campaign can be checked the same way.
Acceptance test reviewDraft · not saved
Engineer: Which units are drifting from the baseline?
Sift AgentsSifting…
Sift Agents: Units 118 and 131 are drifting. Output current has risen about 4% over their last 20 cycles; the other 38 units stay within the baseline.
Output current vs baselineUnits 101 to 140
20 cycles agoLatest
Units 118 and 131Other 38 units
Engineer: Show me the current trend for units 118 and 131 over the last 20 cycles.
Sift AgentsSifting…
Sift Agents: Both units rise about 4% in output current across the last 20 cycles. The other 38 units stay within the baseline.
Units 118 and 131 vs baselineLast 20 cycles
20 cycles agoLatest
Engineer: Draft a monitoring check for these units. Leave the threshold for review.
Sift AgentsSifting…
Sift Agents: Drafted a check for units 118 and 131. The threshold is left for engineering review, and the check stays off until approved.
Output current drift checkDraft · off
Engineer: What changed before this anomaly?
Sift AgentsSifting…
Sift Agents: Reaction wheel 2 speed began oscillating 40 s before the anomaly, right after a commanded mode change. No other channel left its normal range in that window.
Reaction wheel 2 speedBefore the anomaly
No other channel left its normal range
Engineer: Show me the mode change and wheel 2 speed leading up to the anomaly.
Sift AgentsSifting…
Sift Agents: First the commanded mode change, then reaction wheel 2 starts oscillating, 40 s before the anomaly at 14:02:07 UTC.
Event sequence · reaction wheel 214:02:07 UTC
No other channel left its normal range
Engineer: Draft an investigation report with this sequence and supporting evidence.
Sift AgentsSifting…
Sift Agents: Drafted a report that links the sequence to its evidence. It records the timing; it does not confirm a root cause.
Reaction wheel 2 anomaly reportDraft · not saved
Your data. Your engineering judgment.
Give your team the tools to investigate deeply, inspect the evidence, and turn findings into work they can reuse.
Purpose-built tools connect your assets, runs, channels, and existing checks. Agents use Python and SQL to analyze behavior and build visualizations you can inspect.
Follow the investigation from the relevant runs and channels to the analysis and supporting evidence. Inspect the result in context, then ask the next question.
Agents inherit your Sift permissions. Review the evidence, steer the investigation, and approve proposed changes before they are applied.
Inspect the proposed change and its supporting evidence before approving it. Agent access follows your existing Sift permissions.
Flag pressure that remains outside the expected range after startup.
Evaluated during mainstage92 to 104 bar · 0.5 s
00:0000:07.000:10
Run 6 · 00:08.1 to 00:09.0
Startup cleared.Run 6 deviation flagged.
Sift Agents
The check excludes startup and flags a sustained pressure deviation in Run 6.
Turn findings into reusable rules, calculated channels, and reports. Save and share the investigation so the next engineer can continue from the work already done.
Turn a useful finding into a rule, calculated channel, or report. Save and share the investigation so another engineer can continue from the evidence and work already completed.
Focused context and purpose-built analysis tools help avoid unnecessary token usage, so your team can explore more questions while keeping investigation costs in check.
Focused context and purpose-built analysis tools help avoid unnecessary token usage. Engineers can follow the evidence and ask additional questions while keeping investigation costs in view.
Work through a real engineering investigation with Sift. Inspect the evidence, explore the result, and see what your team can carry into the next test.
Already use an AI agent?
Prerequisites
Install the Sift CLI and configure a profile. The MCP server reads its endpoint and API key from that profile.
Sift CLI setupCreate an API key
Confirm the profile works: sift-cli ping
Install the integration
sift-cli agent installDetects each supported client and installs the Sift MCP registration for it. Access is read-only by default. Reload or restart the client afterward.
Find relevant hardware data, compare runs, investigate unusual behavior, perform calculations, and generate visualizations. Agents can also help create rules, calculated channels, and reports, with engineers reviewing proposed changes.
Yes. Agents use Python and SQL to investigate questions that need custom calculations or comparisons. Describe what you want to understand, then inspect the analysis and refine it through follow-up questions.
Agents work with your Sift assets, runs, channels, rules, and annotations. This connects the measurements to the hardware, test history, and existing review work behind your question.
Yes. Save and share investigations, and turn useful findings into rules, calculated channels, and reports. Your team can apply that work to other runs and build on it as testing continues.
Sift uses focused context and purpose-built analysis tools to retrieve and process relevant data. This helps limit unnecessary information in the conversation and avoid unnecessary token usage.
Sift Agents and Sift MCP are available. Request access to use Agents in Sift, or connect your existing AI tools through MCP.
Request Agents Access using this form. Our team will help you get started and confirm the right setup for your program.
Sift Agents runs in your Sift environment. Discuss GovCloud, VPC, or on-premises requirements with our team to confirm which options fit your program. Sift MCP runs on your machine; what reaches an external model depends on your agent and configuration.
