Native SQL for Ontologies: OntoL's Intermediate Library vs Palantir's Approach
The article argues that SQL support for ontology products is a foundational design choice, not a late addition, detailing OntoL's architecture using a relational intermediate library (MySQL/Memgraph) for native SQL execution, a custom DSL for cross-source reasoning chains, and key differences from Palantir's platform-centric approach regarding data residency, query portability, and versioned business logic.
On November 12, 2026, Palantir announced general availability of Ontology SQL and SQL Functions. The author, who works on OntoL, argues that SQL support for ontology products is not a feature added later but a fundamental design choice from day one.
Where Ontology Products Fail
The hardest part is not building the model but making it usable after it's built. Graph query languages like Cypher or SPARQL are a barrier for business analysts; a finance colleague will not learn syntax just to query an overdue receivable. Every new request then falls back to engineers: business files a ticket, someone writes the graph query, schedules it, and delivers a report. The promised "self-service analytics" turns into engineering workload within months. The model's value appears not on modeling day but when the 101st question arrives — can the business side answer it themselves? If every question requires an engineer, the ontology is just an expensive data dictionary.
Two Paths to SQL Support
There are two ways to add SQL to an ontology product:
SQL gateway over graph storage: Translate SQL to Cypher. This sounds elegant but breaks down with aggregations, window functions, and multi-table joins. Performance becomes unpredictable, error messages are opaque to business users, and the translation layer accumulates semantic debt.
Relational intermediate library: Land the ontology on a relational store (MySQL 8 and Memgraph in OntoL's case). Objects, relationships, and attributes have concrete table schemas, indexes, and version fields. SQL runs natively without translation. Structured objects and analytical queries go to MySQL; multi-hop, path, and subgraph matching go to the graph store. Parameters pass through from the upper layer, SQL executes on the relational side, and the DSL takes over for graph traversals.
A critical but often overlooked point: the ontology cannot be just a metadata layer floating above systems; it must have a physical landing point. With a landing point, every SQL statement can be traced to a specific object type, relationship, attribute, and time slice. Without it, SQL is just a string of numbers nobody owns.
OntologyL_Cypher: Making Graph Language Understand Business
The intermediate library solves "SQL runs," but not "metrics drift." A single inference chain often spans three sources: ERP delivery notes, CRM contracts, and intermediate-library inventory reservations. If people write ad-hoc SQL each time, the same "allocatable inventory" metric diverges into dozens of versions across scripts within six months. OntoL's custom DSL, OntologyL_Cypher, supports master-sub inference chains, federated queries, and parameter passthrough. Its role is not to replace SQL but to fix cross-source data-access paths inside the ontology as reusable inference chains. An order ID flows down; each source's query and merge conditions are predefined, not improvised. The reasoning engine also deepens with Allen interval algebra for temporal relations and DE-9IM for spatial relations — capabilities that must land on SQL-addressable table structures or they remain slideware concepts.
Differences from Palantir's Design
The author acknowledges Palantir's clear boundaries: Ontology SQL uses RID references in object mode, supports a subset of Spark SQL, pushes complex queries to Spark, and separates read-only SQL Functions from write Actions with permissions and preconditions. Separating read queries from business writes is a long-advocated practice — querying inventory and deducting inventory should not share the same permission gate.
Three factual differences:
Entry point: Palantir's SQL lives inside Foundry; users must enter the platform. OntoL's SQL runs directly on the intermediate library, so any BI tool, reporting system, or DBA client can connect. This enables portability of business logic — rules live in the intermediate library's tables and functions, so switching front-end applications next year doesn't require renegotiating metrics.
Data location: Palantir stores objects in its own platform; data moves into the platform. OntoL's intermediate library can be the customer's own MySQL; data never leaves the premises. In manufacturing, chemicals, and mining, this is not a preference but a hard requirement.
Metric reconciliation: After a function is published, the same business calculation runs on a dashboard, a mobile app, and an agent — all three numbers must match. A mismatch means one endpoint is still running an old rule version. Documentation alone cannot enforce this; function versioning and test cases are required.
Back to the Shortage Order
Warehouse has 12 units, customer orders 10, requires full shipment. The warehouse clerk says it's impossible. The reason is a straightforward SQL query on the intermediate library: 12 in stock, 5 reserved for other orders, 7 allocatable, shortfall of 3. This query runs natively on the intermediate library without translation. Once published as a function, the clerk's mobile app, the sales dashboard, and the agent all call the same function and return the same answer. If an AI assistant only reads total stock (12), it would incorrectly tell sales "shippable today." The number is accurate but the business logic is wrong. SQL's role here is not to make AI write code but to give AI a vetted data-access path that humans have verified.
Three Judgments
SQL is not a low-end substitute for graph query languages. It is the only language business users will actually write themselves, and the only one that requires zero training for finance and warehouse staff.
The ontology's value is not on modeling day but at the 101st question — can the business side answer it themselves? If not, even a beautiful model is a liability.
The real competition is not who supports SQL first; SQL will inevitably appear everywhere. The competition is who can keep shipping rules, overdue definitions, and allocatable inventory logic inside the customer's own database so they survive an application switch.
Palantir's release is worth watching because it validates this architectural direction. For OntoL, this is not catch-up; they have been here all along.
Signed-in readers can open the original source through BestHub's protected redirect.
This article has been distilled and summarized from source material, then republished for learning and reference. If you believe it infringes your rights, please contactand we will review it promptly.
How this landed with the community
Was this worth your time?
0 Comments
Thoughtful readers leave field notes, pushback, and hard-won operational detail here.
