Upload any time-series CSV — satellite telemetry, sensor logs, IoT streams — and get anomaly detection results in seconds. Statistical + ML detectors run automatically. Zero thresholds to set.
Free forever for 1 satellite · No credit card
No configuration, no domain expertise, no manual threshold tuning. Dsremo learns what normal looks like for your data automatically.
Drag and drop any wide-format CSV with a timestamp column. Satellite telemetry, sensor logs, IoT time-series — any numeric columns work. Auto-resampling handles irregular intervals.
timestamp, battery_v, temp_c, current_mA
2024-01-01T00:00:00Z, 3.7, 22.1, 412
2024-01-01T00:01:00Z, 3.72, 22.3, 409
STL decomposition removes seasonal patterns first. Then 6 statistical + 5 ML detectors analyze every channel. An ensemble vote produces a single confidence score per anomaly.
Every anomaly comes with a human-readable explanation, a severity level, and the detector fingerprint. Mark false positives to improve future detection. Get webhook alerts on critical events.
CSV upload for historical analysis. Live connectors for real-time monitoring.
Upload any wide-format CSV. Auto-detects timestamp column and data frequency. Max 100 MB on Pro.
FreePOST telemetry in real time. Supports batch and single-point ingest. Webhook alerts on anomaly.
FreePull telemetry directly from YAMCS mission planning systems. Paginated, auth-aware.
ProQuery via Flux language. Pulls data for any time range and field set.
ProPull from the open-source ground station network. Works with any public satellite.
ProESA mission data loader and XTCE parameter definition import for standard spacecraft formats.
EnterpriseNo single detector catches everything. Dsremo runs 11 complementary methods and combines their votes into a single confidence score with a plain-English explanation.
Cumulative sum control chart. Detects gradual drift that accumulates over dozens of readings — invisible to point-in-time methods.
Exponentially weighted moving average. Catches sudden level shifts within 2–5 samples of the event.
Classic single-point spike detection. Fast and reliable for impulsive noise and sensor glitches.
Pruned Exact Linear Time algorithm. Detects abrupt structural breaks — when the system regime changes permanently.
Multivariate anomaly detection across all channels simultaneously. Catches cross-parameter covariance breakdowns.
Detects increased noise floor — when a signal's scatter doubles or triples while the mean stays constant.
Trains a tiny GRU neural network on each channel's normal residuals. Flags sequences the model can't reconstruct — nonlinear temporal patterns.
Temporal Convolutional Network with dilated causal convolutions. Captures long-range dependencies across 512+ timestep windows.
Monitors the acceleration of the STL trend component. Early onset detection — catches drifts before they become critical.
Finds subsequences that have never appeared before in the history of the channel. Pure NumPy, O(n log n) FFT implementation.
Monitors rolling Pearson correlations across all channels on a satellite. Flags when one channel decouples from its peers.
No trial periods. No hidden fees. Free plan is genuinely useful for solo operators and researchers.
For researchers, students, and personal projects.
For satellite operators and small teams running real missions.
For mission control teams that need collaboration and audit trails.
For agencies, satellite manufacturers, and constellation operators.
Battery voltage, temperature, attitude control, power subsystems. Self-calibrates per orbital period.
Valve sensors, pressure readings, motor current. SKAB benchmark: F1=90.9% on real valve failure data.
Water quality, power grid telemetry, HVAC systems. Any CSV with a timestamp column works.
Run against standard benchmarks (SKAB, GECCO, OPS-SAT-AD). Export detection results via REST API.
Wide-format: one column for timestamp, remaining columns are your parameters. The timestamp column is auto-detected (you can specify it if it has a non-standard name). Any ISO 8601 or Unix timestamp format is supported. Example:
timestamp, battery_v, temp_c
2024-01-01T00:00:00Z, 3.7, 22.1
Dsremo uses a self-calibration phase on the first 50+ samples of each channel to learn the normal distribution. It sets CUSUM and EWMA control limits automatically from the learned mean and standard deviation. STL decomposition removes the seasonal/orbital component first so your normal oscillations don't trigger false alarms.
Statistical detectors (CUSUM, EWMA, Z-Score, PELT, Isolation Forest, Variance) are fast, interpretable, and work well with as few as 50 samples. They catch mean shifts, spikes, variance changes, and structural breaks.
ML detectors (GRU, TCN, Matrix Profile, Correlation Graph, Trend Velocity) learn complex temporal patterns that statistics miss — nonlinear relationships, long-range dependencies, cross-channel coupling. They need 60–200 samples to train and are available on the Pro plan and above.
Yes. Each account is a fully isolated tenant in the database via PostgreSQL Row-Level Security (FORCE RLS). Your data is never mixed with other customers' data. API keys are stored as SHA-256 hashes — the plaintext is shown once and never stored. All traffic is TLS-encrypted in production. We never log request bodies or full API keys.
Yes. Dsremo has a full REST API with OpenAPI documentation at /docs. You can push telemetry in real-time via POST /api/v1/telemetry, query anomalies, and receive real-time events via WebSocket. API keys are available on all plans.
No. Sign up with Google and start immediately. No credit card required until you upgrade to Pro or Team.
Upload a CSV and get your first anomaly report in under 30 seconds. Free forever for 1 satellite.
Start free with Google