From raw spreadsheet to a working model, in one app.
Local-first AutoML is machine learning that runs entirely on your own hardware. Data, training, and models never leave your network. Cogentic packages it as a zero-code desktop app: upload a CSV, Excel, or Parquet file and it profiles, cleans, trains, and predicts, offline or air-gapped.
Running today on a Fortune 100 production line
Upload, profile, train, predict. Then we tailor and deploy it.
Every frame below is a live recreation of the app. The sidebar switches screens, the buttons run. Figures are captured from a real session.
Drop in a file, nothing leaves your machine.
Bring a CSV, Excel, or Parquet file, including multi-gigabyte exports. It opens in the app on your own hardware; nothing is uploaded anywhere.
- CSV, Excel, and Parquet, into the multi-gigabyte range
- Live preview of every column the moment it lands
- No cloud account, no upload, no telemetry
Dashboard
Explore datasets and preview your data before modeling.
Interactive: tap the sidebar to switch screens
See every column, then fix it with a guided plan.
Cogentic profiles every column on arrival: types, distributions, gaps, outliers. Guided cleaning turns each fix into a versioned, re-runnable plan, so datasets stay reproducible.
- Types, distributions, and missing values flagged up front
- Cleaning wizard records each choice as an auditable step
- Versioned datasets you can re-run on tomorrow's export
Cleaned Datasets
Silver parquet files ready for analysis.
Train a model without writing a line of code.
Pick the column to predict. Cogentic trains and tunes CatBoost, XGBoost, and AutoGluon, promotes the champion, and reports every metric against a held-out test set in plain English.
- Classification or regression, chosen automatically
- Multiple engines compared, champion promoted
- Every run traceable to a re-runnable plan
Trained Models
Compare runs, inspect metrics, and promote a model to inference.
Put the model to work on data it has never seen.
Run the champion on a fresh file for per-row predictions with confidence scores. Charts and the Action Center turn the output into ranked next steps, and flag when a model is too weak to act on.
- Batch predictions with per-row confidence scores
- Ranked drivers and plain-language guidance in the Action Center
- Export to CSV or back into your systems, inside your network
Inference
Run a trained model on new data and review predictions.
The last step is ours: tailored and deployed to your operation.
Software alone doesn't survive a factory floor. We pilot Cogentic on a dataset from your operation, install it inside your network, and tune it to your processes, then train your engineers to run it without us.
Pilot on your data
We run the full path on a dataset from your line and validate the model against your own held-out data before you commit.
Install in your network
Deployed on your machines: a single workstation or an on-prem server, offline or fully air-gapped. Nothing crosses the boundary. Deployable within your ITAR, CMMC, HIPAA, SOC 2, and GDPR controls.
Tune, then hand over
We tailor ingestion, cleaning defaults, and reporting to your operation, then train your engineers to run it themselves.
Six capabilities, one continuous path.
Profile, clean, train, evaluate, predict, visualize. A quality or process engineer does what used to take a data-science team, no code required.
Data profiling
Drop in a CSV, Excel, or Parquet file. See types, distributions, missing values, and a live preview of every column.
Guided cleaning
A wizard fixes types, ranges, and missing values, recording each choice as a reproducible, auditable plan.
AutoML training
Pick what to predict. Cogentic trains and tunes models with CatBoost, XGBoost, and AutoGluon, then picks the champion.
Plain-English evaluation
A leaderboard explains every metric in one line. Know exactly how good a model is without a statistics background.
Inference
Run the trained model on a new file for predictions with confidence scores, ready to export into your systems.
Visualizations
Charts bring data and predictions into one view, and the Action Center ranks what to investigate first, with a flag when a model is too weak to act on.
Questions engineers ask before they buy.
See Cogentic predict on your data.
We'll run the full path on a dataset from your operation: profile, clean, train, predict. Your machine, one session.
Figures in the tour above are from a real session