The data and AI platform you own.
On any cloud, on‑prem or air‑gapped.
Build features, train and serve models, run agents and watch every job in one place, on open Spark and Delta Lake, each run explained, all inside your boundary.
Sovereign AI
Bring the AI to the data. Never the reverse.
Your data, models and agents run inside your boundary, on one governed platform. Nothing leaves unless you decide it should.
Deploy anywhere
Cloud, hybrid, on-prem or fully air-gapped, from one codebase.
Model custody
Your data, your models, your weights: trained and served inside your boundary.
Zero trust by design
Identity-bound access, least privilege and an audit trail for every action, by people and agents alike.
Open by design
Run any model or agent, ours or yours, and swap either without re-platforming.
Where it already runs
- Cloud to air gapSame codebase, zero external dependencies.
- 6 weeksFrom air gap to secure, operational use at U.S. federal agencies and commercial enterprises.
- US · UK · EUDeployments.
- Since 2014In production.
Accreditations
Platform01
One platform, end to end.
Connect your sources, build pipelines and models, deploy agents, watch every run and govern access, without stitching tools together. The same codebase runs in a public cloud region and on a classified network.
- In production since 2014 with commercial and public-sector teams
- Open by design: Apache Spark and Delta Lake on your storage, so tables stay standard
- One sign-in, one permission model and one audit trail for people and agents
- 01Data ReadyIngest, resolve and structure at petabyte scale.
- 02CoreGoverned processing, identity and orchestration.
- 03AI ModelsPredictions trained on your full history.
- 04AI AgentsAgents that decide and act within your rules.
- 05AudiencesActivation to every channel you use.
Capabilities
Industry-leading capabilities,
production-grade from day one.
Ten capabilities that take AI and data work from first query to governed production, on one platform you control.
Build01
Draw the pipeline.
Code where it helps.
An AI pipeline on the visual canvas: a feature build in custom PySpark, a GPU-trained model registered in MLflow, batch scoring and an audience, fed by sessionized web events. Follow every run in Operations, or test one step in Interactive mode.
Learn more about workflows and connectorsProcesses
ConnectionEvent StoreID StoreSpark processorSpark MLMLflowPythonDelta outputDashboardAudience Agent+ 44 more
| person_id | sessions_30d | avg_order | churn_risk |
|---|---|---|---|
| p_10482 | 14 | 84.20 | 0.12 |
| p_10517 | 3 | 41.00 | 0.71 |
| p_10533 | 22 | 126.75 | 0.05 |
| p_10561 | 7 | 58.10 | 0.38 |
| p_10602 | 1 | 19.99 | 0.83 |
All runs
Every run in Development. Select a run to see its steps.
Notebooks & data science02
Notebooks your whole
team can work in.
Train a model, log it to MLflow, chart the scores and check for drift, in JupyterLab with Python and Scala kernels (each with a GPU edition). Spark starts inside the notebook, then attaches to a cluster when the data grows.
- RAN BY stamps: who ran each cell, when, for how long and where
- Real-time co-editing with your team
- Start on local Spark, then attach a remote cluster
Illustrative people and data · modeled on the Syntasa notebook
Agentic AI03
Agents that do the work, inside your rules.
A Super Agent takes a goal, briefs specialist agents and shows every step and every reason. Agents use the same roles, service accounts and audit trail as your team, on the model you choose, including self-hosted.
Customer Insights
Ask your data; get answers with the query behind them.
Customer Audience
Describe an audience; get a segment checked on live data.
Engineer
Code search, dependency analysis and reviews.
CloudOps
Daily cost tracking with anomaly alerts.
Geospatial & visualization04
Query, index and map location data at scale.
GeoMesa spatial SQL is registered the moment Spark starts in a notebook. H3 and Apache Sedona libraries are in the runtimes. Index points into hexagons, roll them up by region, or search within any area you draw.
- H3 hexagon indexing at the resolution you choose
- GeoMesa spatial SQL, ready in every notebook
- Apache Sedona for spatial joins at scale
- Map dashboards: deck.gl layers, choropleths and heatmaps
Observability · Lens05
Every run, explained.
When a job or notebook fails, Lens shows what happened, why, and what to change, in plain words. It also shows the compute you wasted, in core-hours.
OverviewKPI tiles, runs per day, where the time goes and why runs fail.
ActivityEvery run, filterable, with the likely cause and top recommendation beside it.
JobsExecutions by job, average job time and last-run health.
InfrastructureFleet health, capacity and autoscaling, with idle warnings.
Causal traceFrom the failure, layer by layer, to the root cause and the fix.
NotebooksWorkspaces and notebook runs, local or on a remote cluster.
Performance06
Lakehouse tables that stay fast.
Delta Optimize keeps tables compact and well organized in planned, resumable runs. The Metadata Accelerator answers many queries from table metadata, so they read far less. Delta Track shows table health over time.
- Bring existing tables under management, partition by partition
- Incremental: only partitions that changed are re-optimized
- Compaction, Z-ORDER and liquid clustering on open-source Delta Lake
- Time saved (modeled) and files compacted, shown in Lens
Dataset healthEvery managed table scored, with file shape and layout.
File trendsFiles, sizes and health over 30 days, table by table.
Optimize runsLiquid, Z-ORDER and compaction runs, files before and after.
Managed tablesTables brought under management and their baseline.
Metadata AcceleratorAccelerated queries, time saved (modeled) and scan avoided.
Compaction
Many small files become a few right-sized ones.
Table health 58 before, 96 after.
Z-ORDER
Rows sorted across files by the columns you filter on, so queries skip more files.
Table health 61 before, 94 after.
Liquid clustering
Data grouped into clusters that adapt as the table grows.
Table health 57 before, 97 after.
Scale07
From one notebook to
thousands of nodes.
Apache Spark 3.5 and Delta Lake 3.3 with autoscaling workers, GPU runtimes, Kubernetes containers and Kafka streaming. Capacity follows the work, up and back down.
What that looks like in the product: every write, merge and optimize recorded in the Delta table's history, workers added and removed as load changes, and Lens counting the runs and core-hours.
Delta table history customer_features
| Version | Timestamp | Operation | Rows | Files |
|---|---|---|---|---|
| 412 | 2026-10-08 14:30 | write | +1.2M | +48 |
| 411 | 2026-10-08 14:00 | merge | +310K | +22 |
| 410 | 2026-10-08 13:30 | write | +1.1M | +46 |
| 409 | 2026-10-08 13:00 | optimize | 0 | −296 |
| 408 | 2026-10-08 12:30 | write | +1.3M | +51 |
Autoscaling spark-large
Lens last 14 days
Delta table historyEvery write, merge and optimize on the record.
Lens · executions by jobRuns and job time across the platform.
Petabytes
of data on open Delta Lake
Billions
of events every month
Trillions
of historical records processed
Thousands
of nodes, scaled to the workload
Flexibility08
Your cloud, your data center or no network at all.
Four deployment models from one codebase. Run on Amazon EMR, Google Dataproc, Azure HDInsight, Kubernetes, Docker containers or your own Spark, with the same screens and security model everywhere.
Cloud, hybrid, on-prem or air-gapped. Hybrid simply means some workloads run in your cloud and some in your data center, by classification.
One codebase · same screens, features and security model
Amazon Web Services
Your AWS account and region.
Google Cloud
Your project, region and keys.
Microsoft Azure
Your subscription and region.
Your data center
Your hardware and network.
Air-gapped
No route to the internet.
Usability · newly redesigned09
Calm, clear and made to fit you.
A redesigned experience with eight themes, eight accent colors or your own, and automatic light and dark. The theme carries into notebooks, dashboards and ML tracking.
- Light, Soft light, Warm light, Light with dark sidebar, Dark, Nord, Carbon and High contrast
- Eight accent colors, or your own
- Favorites, saved filters and folder-wide search
- Trash with restore, and searchable help with animated how-tos
Favorites
Star what you use most.
Search everywhere
Folder-wide search across apps and notebooks.
Trash and restore
Deleted by mistake? Bring it back.
Saved filters
Your views, one click away.
Help and shortcuts
Searchable help with animated how-tos.
Live run status
See every node's state on the canvas.
Runtime schedules Sample data · this week
Cost control10
Spend on work, not on idle.
Runtimes start and stop on a schedule, shut down when idle and can terminate when a job finishes. Lens shows wasted compute so you can fix it at the source.
- Daily, weekly, monthly or cron schedules
- Maintenance-window pause that restores every schedule exactly
- Runtime cost estimates, with offline price lists for disconnected sites
- Fewer small files and faster queries mean less compute
How we are different
Built for the data you can't send anywhere.
The full platform, air-gapped
Not a reduced edition. The same platform installs from offline bundles, with government sign-in.
Every run explained
Lens traces each job and notebook run from failure to root cause to fix.
Visual and code, one canvas
Analysts drag and drop; engineers open the same step in SQL, Python, R or Scala.
Built for behavioral data
Event Store and ID Store turn raw events into sessions, visitors and an identity graph.
Open, with no lock-in
Apache Spark and open-source Delta Lake on your storage. Zero-copy to the warehouse you have.
One codebase everywhere
Public cloud to classified networks. Accreditation travels with the platform.
Architecture
From raw data to decisions and actions.
Five modules, one pipeline, all inside a perimeter you control. The architecture is the same whether it runs in a commercial cloud region or on a network with no route to the internet.
- InputYour dataEvery source and format, as it is.
- Module 01Data ReadyIngest, resolve and structure.
- Module 02CoreGoverned processing, identity, orchestration.
- Modules 03–04AI models & agentsModels predict; agents decide and act.
- OutputDecisionsAudiences, workflows, dashboards.
Deployment substrate
Customers
In production with teams who can't compromise.
Global brands and public-sector teams, plus agencies across the U.S. national security community whose work is not described publicly.
- U.S. National Security Organization
- Major Research University
State of Arizona
State of California- State Department of Transportation
- U.S. Regulatory Agency
“Generated millions of dollars in incremental business value.”
See your data and workflows in Syntasa.
Bring a real constraint: a network that cannot connect, data that cannot move, or a pipeline that keeps failing. We will show you how it runs on Syntasa.