AI that turns questions into queries, on every engine you run.
Ask in plain language. Compass AI finds the data, writes the query, tunes it, picks the engine, and waits for your approval before anything runs.
Runs in your environment with the AI model of your choice. Holds metadata only.
WITH last_night AS (
SELECT COUNT(DISTINCT user_id) AS logins
FROM core.users
WHERE dt IN ('2026-10-07','2026-10-08')
AND login_ts BETWEEN ... ),
yoy_night AS ( ... )
SELECT 'last_night' AS win, * FROM last_night
UNION ALL
SELECT 'yoy_night', * FROM yoy_night;
One question, four things to figure out first.
Before anyone can write the query, they have to answer these. Every time.
Where is the data?
Spread across catalogs, databases, and plain files.
Is the schema current?
Columns change. Views hide behind table names.
Which engine, and when?
Not every engine reaches every table. Not every hour is quiet.
What settings?
Memory, parallelism, engine flags. Wrong, and it fails or starves production.
Compass AI answers all four, and keeps a person in control of the last step.
How the AI works.
The AI does the thinking. You do the publishing. Every step is logged.
Ask
YouType your question. Optionally limit which engines may run it.
AI understands the request
AIThe AI works out what you are asking for: metrics, time windows, entities.
AI writes the query
AIThe AI matches entities to your real tables and files, then writes SQL or DataFrame code, whichever the engine needs.
AI tunes and routes it
AIA second AI pass tunes the query, proposes an engine and a time window, and picks settings from what that engine allows.
You approve
YouThe AI proposes; it never runs anything. You review, adjust settings, and publish or reject.
AI learns from the result
AIThe outcome is recorded and makes the AI better on the next request.
The AI proposes. You decide.
Pick the engines you would allow and watch the AI route the query. Edit a setting out of range and watch it get rejected.
Engine scope
AI, grounded in your data.
The AI only sees what the inventories tell it, so it answers about your estate, not a generic one. Runs on Kubernetes, VMs, or bare metal.
Inventories
- Service inventory what is installed, and where
- Data inventory what data exists, and in what shape
- Job-history inventory what has run, batch and streaming
AI stages
- AI query generation your question becomes a query, grounded in your inventories
- AI optimization and routing tunes the query, picks the engine and the window
Governance
- Approval and publish no AI inside; the only thing that can run a query
- Audit every action, append-only
- Authentication built in, or your identity provider
Operations
- User interface requests, approvals, dashboards
- Observability current load on every engine
- Data map what a query will touch, shown at approval
- Internal database yours, or PostgreSQL if you have none
Plugin layer
Every system is reached through a connector with four operations. Add a new engine by adding a connector; nothing else changes.
- enumeratelist objects
- describeget metadata
- declare vocabularyallowed settings
- submitrun a query
- workload stateoptional: current load
Works with what you have.
No migration. Data stays where it is.
Substrates
Hadoop and Cloudera CDP, Kubernetes, Azure, AWS, GCP, standalone database servers
Table stores
Apache Hive, Apache Iceberg, Snowflake, Databricks
Database servers
PostgreSQL, MySQL, Azure SQL, HBase, Cassandra
File stores
NFS, CEPH, Azure Files, object storage
Engines
Hive, Impala, Spark, Trino, database servers, cloud warehouses
Streaming and history
Kafka, Flink, YARN, Spark history server, engine query logs
Examples, not an exhaustive list. All names belong to their owners.
Licensing.
Compass is licensed, not hosted. The AI runs inside your network, with a hosted model or one you host yourself. Terms per engagement.
Enterprise deployment
For organizations running a heterogeneous data estate
- Unlimited engines and connectors
- Your database, your identity provider
- Custom connectors on request
OEM and embedded
For platform vendors and systems integrators
- Embed Compass AI in your product
- Private connectors
- Co-development for your engines
Pilot
For a scoped evaluation on one estate
- A few engines, fixed scope
- Read-only first, publishing on request
- Converts to a full license
Company
Nivika Technologies builds AI for organizations whose data has outgrown any single platform. Compass is its first product.
The company is founded by Nitin Jakka.
Talk to us.
We reply within two business days.