You declare a logical model
Define entities, relations and the constraints traffic cannot reveal: atomicity, data residency, personal fields and a cost ceiling. Register the engines available to the application.
Managed Data Layer, Open Core
Smart Data Engine
AI proposes the layout. Your rules set the limits.
Declare entities and business constraints. Smart Data Engine plans placement; its open SDK connects your application directly to the engines. Run databases yourself or use our hosting. The control plane never receives your rows or database credentials.
Early-stage product. Public SDKs are available; the integrated AI workflow, demo and onboarding are still being developed.
How It Works
The first adoption scope is a new application or module, agreed and qualified with your team.
We decide the physical schema
A placement describes the target engine and physical schema. AI proposals are checked against declared constraints and recorded with the evidence behind the decision.
Your application stops naming tables
The SDK reads and writes directly to the engines using local credentials. The control plane is outside that request path, and the application works with logical entities.
Approve, migrate, verify
A placement change needs approval. Data is copied and verified locally before cutover. We qualify the rollout and its pause budget against the selected workload.
Why Smart Data Engine?
Because the physical schema is a decision somebody has to keep making.
No expiry in the SDK
Ending the subscription stops new placement maps, not the SDK. The application can keep using its last accepted placement. Database hosting is a separate service and agreement.
AI proposals, explicit safeguards
AI components propose layouts, migration plans and analyst queries. Deterministic checks reject constraint violations and record the decision. AI is never the judge of whether a migration copied the data correctly.
A data boundary you can inspect
The Apache-2.0 client libraries are public. You can inspect the code that applies placement maps and connects to databases. Rows and database credentials do not enter the control plane.
One group, one engine
Entities that must commit together stay in one engine and use its transaction semantics. There are no distributed transactions or cross-engine joins hidden behind the abstraction.
Verified migrations
Migration components include resumable copying, value-level verification and controlled cutover. We agree and test the rollout, including its allowed pause budget, against your workload.
The Way Out Is Documented
Export the physical layout and access your data with the engine's native driver. The documented exit path also explains the limits, so you can assess them before adopting the SDK.
Planned analyst API
A query and the code to run it
Ask for a metric in plain language and choose a programming language. The planned API will return the query, its target engine and a ready-to-use read program based on your declared model.
The program runs in your environment. Results and database credentials stay there.
Example request
Give me a query to read PnL for the last 7 days.
Query and target engine
SQL, parameters and the engine selected from your placement map.
A ready-to-use read program
Code in your chosen language, with installation and local configuration instructions.
Query-generation components already exist. API access and program generation are planned.
You Do Not Have to Trust Us. Read the Code.
Read the open client libraries to inspect routing, signed maps and the data boundary. Our public L2 orderbook engine is separate engineering evidence, with source, tests and benchmark methods you can reproduce.
- C++20 Native L2 orderbook engine
- Open Source under Apache-2.0
- Measured Public benchmark methods
Results depend on hardware and workload. The public comparison documents its methods and limitations; it is not a performance guarantee for your application.
Read the client libraries (opens in a new tab)Read benchmark methods and results(opens in a new tab)
The Python library on PyPI(opens in a new tab) The TypeScript library on npm(opens in a new tab) Which languages exist today, and what each one supports(opens in a new tab)
Where the Boundary Is
Your SDK connects directly to the engines. The control plane receives model metadata and aggregated telemetry, not database rows or credentials. Its operational controls and current limits are available for review.
-
Not one row crosses
Placement uses the declared model, operation shapes and aggregated telemetry. Data values and database credentials stay outside the control plane. The client code enforcing that boundary is public.
-
Even the names can be hashed
Sensitive entity and field names can be replaced with keyed identifiers inside your infrastructure. The key stays local, while the control plane works with the resulting model metadata.
-
Review the controls, not a badge
We do not claim SOC 2 certification. Before production, review the agreed access model, signing-key handling, audit records, recovery procedures and workload qualification with us.
-
An old map keeps working, a replayed one does not
The SDK verifies signed placement maps and refuses older versions once a newer one has been applied. Existing accepted maps do not expire when the control plane is unavailable.
Your Engines, Wherever They Already Run
Keep engines on your servers, inside your VPN or in a cloud you already use. Hosting by us is scoped separately with your team. In either case, the SDK connects directly to the selected engines.
How We Start
Agree the scope before committing to an integration.
-
Share the workload
Tell us about the application, access patterns, constraints and available engines. A new module is a clearer first scope than retrofitting an entire existing system.
-
Agree what to prove
Define the data model, required operations, cost assumptions and acceptance criteria. We state what the current libraries support and what remains to be built.
-
Validate the integration
Check local credentials, signed maps, reads and writes on the selected engines. Verify recovery and operational ownership before any production rollout.
-
Review changes with evidence
Use observed traffic as a baseline. Any placement change needs an explained proposal, approval and a migration process qualified for your workload.