Apivom Abu — an AI engine that turns questions into safe queries
Abu is the AI engine behind the Apivom products. It understands a question in plain language, finds the right context, writes a query, checks it and runs it read-only. Abu is included with the products and is not sold separately.
Infrastructure layerAbu in numbers
Natural language to data, with the checks built in
Questions, saved views, alerts and digests all go through one engine, so the AI and query logic lives in one place and the apps stay thin.
Question to query
Abu reads a plain-language question, retrieves the relevant schema notes, examples and vocabulary, and writes a query for one customer’s data.
Two-layer guardrails
Every query is checked before it runs, then executed read-only with a time limit and a 1,000-row cap.
Semantic models
Common questions compile into a fixed query from a semantic model, so a saved view runs again without calling AI.
Your vocabulary, learned
Glossary terms and approved examples are kept for each customer and retrieved when a question comes in.
Admin console
Seven pages to curate schema docs, examples, glossary, system prompts, retrieval settings, feedback and sources.
Scheduled sweeps
Alerts and digests are evaluated hourly, and edited knowledge is re-embedded daily, so answers follow your content.
How the Apivom products use Abu
You do not set Abu up on its own. It is the engine the products call when a question has to be understood, an insight has to be saved or a sweep has to run. Here is where it works today.
- Pivot. Analysts ask in plain language and get the query, a table and a chart. Abu understands, checks and runs it read-only; Pivot stays a thin app.
- Iris insights. Published views appear in Iris as dashboard widgets. Abu evaluates alerts hourly and prepares daily and weekly digests, each recipient seen with their own scope.
- Iris ticket triage. Abu summarises and labels a support ticket and can consult vendor documentation. It never writes to Iris data; Staff stores the result.
- Staff. Staff starts ticket triage and runs the scheduled work: the hourly sweeps and the daily re-embedding are Staff schedules.
Guardrails that run on every query
A language model writes the query, but it does not get the last word. Every query passes a static check, and then runs under a read-only role that cannot change anything.
- Allowlists. Only a single read query. Every table must be schema-qualified and allow-listed, functions are allow-listed, and system catalogs are off limits.
- Read-only execution. A read-only transaction, a time limit and a 1,000-row cap. A result that was cut short is flagged as truncated.
- One customer at a time. Each customer has its own read-only access role, and the customer organisation comes from the signed-in session, not from the question.
Frequently asked questions
What is Apivom Abu?
Abu is the AI engine behind the Apivom products. It turns a plain-language question into a checked, read-only query on one customer’s data, and it powers saved views, alerts, digests and support-ticket triage. It also has an admin console for the knowledge it relies on.
Is Abu sold separately?
No. Abu is an infrastructure layer that is included with the Apivom products. You do not buy or set it up on its own. To see what it does for your use of Pivot or Iris, talk to us.
How does Abu keep queries safe?
A query is checked before it runs: one read query only, tables must be schema-qualified and on an allowlist, functions are allow-listed and system catalogs are blocked. It then runs read-only with a time limit and a 1,000-row cap. Published semantic views can also be scoped per viewer: own records, team or company.
How does Abu learn our vocabulary?
Through content you control. Admins curate a glossary, schema notes and approved example questions in the console, and Abu retrieves them for each question. A thumbs-up answer can be promoted into an approved example. Edited content is re-embedded daily.
Where does our data go?
Queries run read-only on your own data, one customer at a time. Conversation history keeps the question and query definitions, not result data, and the view assistant suggests a view without sending rows to the model. Features that explain a result send a limited sample of rows. For hosting and processing details, talk to us.
Which Apivom products use Abu?
Pivot uses it for its question console. Iris uses it for insights: saved views, dashboard widgets, alerts, digests and support-ticket triage. Staff starts triage and runs the scheduled sweeps. Other products do not use Abu today.
Answers from your data, with the checks built in
Abu comes with the Apivom products. Talk to us about the questions you want to stop asking an analyst.