Governed semantic layer

Stop building semantic layers by hand.

Replace months of manual semantic engineering with deterministic execution built from approved business knowledge. No LLM-generated SQL. Every answer is explainable, governed, and fully auditable.

  • Deterministic execution
  • No LLM-generated SQL
  • Human-approved business knowledge
  • Complete execution trace

Every answer, explained

“Top customers by revenue, last quarter”

Matched approved business intent: Customer revenue ranking

Executed the approved query template, read-only

Resolved “last quarter” to explicit dates, shown in the trace

Confidence: HighFull execution trace available

Technology partner

QuantumLayers

Built for deterministic AI over enterprise data.

Explainable semantic execution for production AI systems.

Supports SQL Server, PostgreSQL, MySQL, MariaDB, Oracle, Snowflake, Databricks, ClickHouse, Google BigQuery, and Amazon Redshift.

What Smundar is

A semantic layer that understands your business and proves it

Questions execute against approved, deterministic query templates. A language model never writes SQL against your data.

Understands

Smundar maps your schema to your business terminology (customers, orders, revenue), not table and column names.

Governs

Smundar proposes; your team approves. Nothing answers questions until a person has signed off.

Explains

Every execution records what was selected, what was rejected, and how confident Smundar was.

How it works

From connection to trusted answers in four steps

  1. 1

    Connect, read-only

    Read-only credentials, encrypted at rest, scoped to approved tables. Revocable at any time.

  2. 2

    Map your terminology

    Smundar analyzes your schema and proposes mappings to your business terminology.

  3. 3

    Approve the knowledge

    Your team reviews every dataset, mapping, and business intent. Only approved knowledge can answer.

  4. 4

    Ask anywhere

    App, API, or any MCP client. Every answer is deterministic and explained.

The product

What working with Smundar looks like

Semantic Query

Ask in plain language

Type the question the way your business asks it. Smundar matches it to approved knowledge. No SQL and no schema knowledge required.

app.smundar.com/query
The Explore Datasets page showing the question “show customers” matched to the Customers dataset with a Good match confidence badge and a reasoning summary.

Explainability

See why an answer was chosen

Every answer opens into its execution trace: the dataset selected, the candidates rejected, resolved parameters, and a confidence label.

app.smundar.com/governance/executions
An execution's explainability panel showing a full score breakdown with point values and the ranked alternative datasets that were considered and rejected.

Business System Resolution

The right system, every time

When more than one connected system could answer, Smundar resolves the right one or asks you to choose. It never silently guesses.

app.smundar.com/ask
The Ask Smundar page showing a Dataset Resolution panel that deterministically resolved the Customers dataset — a single clear winner, not a guess.

Governance

Approve what AI can know

Datasets, mappings, and business intents wait for human approval before they can answer. Your team decides what production knowledge looks like.

app.smundar.com/governance/review
The Review Queue page showing suggested datasets Smundar discovered, waiting for human review before they can be queried.

MCP

Any MCP client, governed

Connect MCP clients with scoped API keys. AI agents inherit the same approvals, determinism, and audit trail as everyone else.

app.smundar.com/settings/api-keys
The API Keys page showing the account's scoped API key used by external systems and MCP clients, plus a ready-to-run JSON quick-start example.

Why deterministic AI matters

When AI guesses against your database, a wrong answer looks exactly like a right one

Text-to-SQL demos well and fails quietly. Smundar separates interpretation from execution: models may help understand the question, but only approved, deterministic logic touches your data.

Typical text-to-SQLSmundar
Who writes the queryA language model guesses SQL from your schemaA human-approved, deterministic query template
Same question, asked twiceMay produce different SQL and different answersIdentical execution, identical answer
Wrong-but-plausible answersLook exactly like right answersStructurally prevented: only approved logic runs
When it can't answer confidentlyGuesses anywayRefuses explicitly and asks for clarification
AuditabilityOpaque model outputFull execution trace: selection, rejection, parameters, confidence

Key features

Built for organizations that need answers they can stand behind

Deterministic execution

The same question always runs the same approved logic. Answers are repeatable, testable, and defensible.

Explainable results

Every answer shows what was selected, what was rejected, how parameters were resolved, and why.

Governed semantic layer

Business knowledge lives in a reviewed, versioned layer your team controls, not in a prompt.

No hallucinated SQL

Language models never write SQL against your database. Only approved query templates execute.

Human-approved knowledge

Datasets, mappings, and business intents pass an explicit approval workflow before they can answer anything.

Enterprise security

Read-only access, encrypted credentials, role-based access control, and a full audit trail of every execution.

Multi-connector support

Ten databases and warehouses, from SQL Server to Snowflake, behind one consistent semantic layer.

MCP for AI agents

AI agents connect through the Model Context Protocol and inherit the same governance as every other caller.

Supported connectors

One semantic layer across ten databases and warehouses

Connect the systems you already run. Every connector gets the same governance, determinism, and explainability.

Security

Enterprise security, without the asterisks

Built for environments where data access must be scoped, governed, and auditable.

Read-only by design

Smundar queries your data; it never modifies your source systems.

Encrypted credentials

Connection credentials are encrypted at rest and never exposed in the UI, API, or logs.

Role-based access control

Owner, Admin, and User roles gate every governance and administrative action.

Full auditability

Every semantic execution is logged with its complete execution trace.

Your data stays yours

Queries execute directly against your database. Smundar is not a data warehouse and does not train models on your data.

Revocable at any time

Rotate or revoke credentials whenever you choose. Access ends immediately.

Read the full security overview →

Proof before purchase

Validate Smundar on your own data. Free.

Before you commit to anything, watch Smundar answer the questions your business actually asks, using your own data.

1

Choose your questions

Select up to 30 business questions that matter most to your organization.

2

Grant safe access

Read-only access to a test, staging, or sanitized environment containing the relevant data.

3

We do the work

We configure Smundar, map your terminology, and validate every scenario at no cost.

Your safeguards

  • • Access is limited to approved tables and read-only operations
  • • Smundar never modifies your source systems
  • • Credentials can be revoked at any time
  • • No obligation, no payment details, no lock-in
Start a validation pilot

If Smundar demonstrates value through deterministic execution, explainable answers, and fast responses, we’ll discuss a production deployment. If not, revoke access and walk away.

Pricing

Straightforward plans, from first agent to full organization

Every plan includes deterministic execution, explainable results, and governed semantic access. Monthly pricing, no usage surprises.

Included with every plan

  • Deterministic execution
  • Explainable results
  • Governed semantic layer
  • Read-only enterprise security
  • MCP integration
  • Multi-database support

Developer

Connect AI to your own business data.

$99USD / month

14-day free trial

Capacity

  • 3 team members
  • 2 database connections
  • 500 queries per day
Recommended

Team

Build and share governed business knowledge across your team.

$499USD / month

Capacity

  • 10 team members
  • 10 database connections
  • 10,000 queries per day

Enterprise

Run production AI on governed enterprise knowledge. Contact Sales.

Custom

Capacity

  • Unlimited team members
  • Unlimited database connections
  • Unlimited queries

FAQ

Common questions

How is Smundar different from text-to-SQL tools?

Text-to-SQL tools ask a language model to write SQL against your schema. SQL is generated for each request, so the result depends on whatever query the model produced that time. Smundar never lets a model write SQL. Questions match human-approved, deterministic query templates: the same question always runs the same logic, and every answer carries its explanation.

Can Smundar modify my data?

No. Smundar operates read-only by design. It executes approved read queries against your database and never writes to your source systems.

What happens when Smundar can't answer confidently?

It refuses and tells you why, asking for clarification or showing which business knowledge is missing. Smundar never silently guesses, because a confident wrong answer is worse than no answer.

What is MCP, and which AI assistants can connect?

The Model Context Protocol (MCP) is an open standard that lets AI assistants and agents use external tools. Smundar exposes its semantic layer over MCP — certified against MCP Inspector, the OpenAI Agents SDK, LangChain, and raw JSON-RPC, with Claude Desktop supported via a local bridge — so any conforming client can query your data with the same governance, determinism, and audit trail as a person using the app.

Which databases does Smundar support?

SQL Server, PostgreSQL, MySQL, MariaDB, Oracle, Snowflake, Databricks, ClickHouse, Google BigQuery, and Amazon Redshift, all behind one consistent semantic layer.

What IP address do I need to whitelist for my database firewall?

145.132.40.28 — Smundar's dedicated static outbound IP for all production database connections. Whitelist it if your SQL Server, PostgreSQL, MySQL, MariaDB, Oracle, MongoDB, ClickHouse, or Redshift database sits behind a firewall or security group. It doesn't apply to Snowflake, Databricks, or BigQuery, which Smundar reaches over the provider's own managed API endpoint rather than a customer-operated firewall.

Do you train models on our data?

No. Your data stays in your database. Smundar's semantic layer stores approved business knowledge (terminology, mappings, query templates), not your records, and nothing is used to train models.

How does the free validation pilot work?

You pick up to 30 real business questions and grant read-only access to a test, staging, or sanitized environment. We configure Smundar, map your terminology, and validate every scenario at no cost. If the results earn your trust, we discuss production; if not, you revoke access and walk away.

Ready for answers you can stand behind?

Start with a single connection, or let us validate Smundar on your own data. It's free, read-only, and comes with no obligation.

Or send us a message directly