Understand the path from question to query.
And verify every deployment boundary.

Spotonix is the AI analyst that interprets before it queries. This page is for the people who have to sign off on that. It separates the stable product path—interpretation, query generation, validation, and execution—from deployment choices such as hosting, warehouse credentials, model credentials, identity, network routing, and retained artifacts.

The product behavior can be demonstrated. The infrastructure boundary must be documented for the selected deployment; this page does not turn engagement-specific choices into universal claims.

How it works

Three architectural responsibilities.

A production evaluation should separate the system's responsibilities clearly: represent the intended analysis, generate and validate a query, and execute through a configured warehouse connection. Each boundary has a different control owner.

01 · Interpretation

Build a visible analysis plan

Intent Algebra binds business terms to the Context Graph and represents the intended Segments, Calculations, filters, grain, and scope as a plan. In Copilot mode, the workflow pauses for plan acceptance before query generation continues.

02 · Generation

Generate with a configured model backend

The current implementation supports configurable Anthropic and OpenAI model backends. The query generator receives the question and the analysis plan. Provider, model, key ownership, and network path are confirmed for the deployment.

03 · Validation & execution

Check bindings, then use the configured connection

Semantic checks compare material query bindings to the accepted plan before execution. SQL then runs through the configured database connection. Whether that connection uses a service identity, delegated user identity, or another credential mode is deployment-specific.

The path a question takes

1

Configure the Context Graph, model backend & warehouse

2

Connect your warehouse & your model

3

Ask · Intent Algebra resolves a visible plan

4

Generate SQL · validate material bindings

5

Execute through the configured database connection

The interpretation and closure steps are the product; see Intent Algebra for how the plan is bound, grounded, and exposed before SQL executes.

What is fixed and what is selected

A diligence table, not an implied topology.

The product path is stable; the deployment topology is not universal. Use this table to drive the architecture review and record the selected boundary for the actual engagement.

Your warehouse

Queried through a configured database adapter

Credential mode, network path, and warehouse controls

The model

Anthropic or OpenAI backend can be configured

Provider, model, key ownership, endpoint, and retention terms

The Context Graph

Provides explicit domain context to interpretation

Storage location, backup, access, and export requirements

The accepted plan

Visible before execution in the analysis workflow

Retention, stable identity, audit export, and access policy

The defensible claim is the visible interpretation and binding validation. Spotonix does not currently claim deterministic plan-to-query compilation or a universal logged-in-user execution model.

Where the boundary is

What this page does not ask you to assume.

An evaluation is easier when unknowns are explicit. Confirm these items in the architecture packet for your deployment instead of inheriting them from generic website language.

  • Do not infer model-data usage terms. Confirm the selected provider, endpoint, retention settings, and contract before sending production context.
  • Do not infer zero retention or zero egress. Document prompts, query results, logs, caches, Context Graph artifacts, and every system that receives them.
  • Do not infer delegated user identity. Record the warehouse credential mode and test table, row, and column access with representative users.
  • Do not infer that every ambiguity blocks execution. Demonstrate the material ambiguities the product surfaces and document which assumptions the workflow permits.

Controls & assurance

What your security team gets to hold.

Identity & permissions

Engagement

Confirm application identity, warehouse credential mode, provisioning, and the enforcement point for table, row, and column restrictions. Runs-as-user is not a universal live claim.

Data residency

Engagement

Confirm where the application, Context Graph, logs, caches, and query results are stored, processed, backed up, and deleted for the selected topology.

Model governance

Live

Anthropic and OpenAI backends are configurable. Confirm provider, model, credential owner, endpoint, regional availability, retention terms, and model-change procedure.

Auditability

Engagement

The analysis plan is visible in the workflow. Stable plan identity, retention, query-to-plan linkage, and export requirements are confirmed during evaluation rather than assumed.

Graph portability

Live

The Context Graph is your asset. Open export of the graph in a standard, non-proprietary format so it leaves with you is in development — we will confirm the format and scope with design partners rather than overstate it here.

Formal certifications

Roadmap

Spotonix does not currently claim a completed SOC 2 attestation. Any compliance or contractual requirement must be reviewed directly before production use.

Deployment topology and control specifics are confirmed per engagement. We work these through with design partners against your real environment, not a reference diagram.