Ontology API

Let agents inspect real structure before they query.

RushDB exposes a live snapshot of project structure as compact Markdown or structured JSON. Applications can inject it into agent context or use it to build controlled interfaces.

Start free View docs

Agents guess when the current data model is missing from context.

A generic prompt cannot tell an agent which labels exist, how fields are named, which numeric ranges are realistic, or which relationship paths can be traversed.

Before

Guess field names from a prompt
Filter outside the available value range
Attempt relationship paths that do not exist
Maintain schema prompt text by hand

With RushDB

Read labels and property types from the live project
Inspect sample values and numeric or datetime ranges
See relationship types and directions
Check vector-index readiness before semantic search

What it enables

Agents plan from the database, not from a prompt guess.

Ontology turns RushDB into a self-describing backend for agents and adaptive interfaces. The first call tells the agent what data exists and what queries are safe to construct.

Grounded tool calls

Agents can read the current data shape before they call search, aggregation, or relationship traversal tools.

Dynamic filters

Applications can build filter chips, numeric sliders, and date ranges from ontology output instead of hardcoding UI options.

Less prompt maintenance

As records arrive and the schema evolves, the agent reloads ontology instead of relying on stale hand-written schema notes.

How it works

Start with the smallest useful path.

Choose Markdown or JSON

Use Markdown for compact LLM context injection. Use JSON when application code needs structured schema data.

Load the relevant labels

Request the whole ontology or restrict the result to the labels needed for the current workflow.

Build queries from real structure

Use the returned fields, values, ranges, relationship directions, and vector-index metadata to constrain the next tool call.

Flow

Schema awareness loop

The docs describe a stateless loop: load ontology at session start, let the agent plan from available structure, then execute a query that matches the current project.

Session starts

Load ontology

Plan valid query

Execute SearchQuery

Use grounded result

Implementation sketch

Load compact ontology context at session start.

The Markdown endpoint is intended for LLM context injection. Pass labels when the workflow only needs one part of the project.

Know the operational boundary.

Ontology is a live snapshot

The response describes labels, properties, value samples or ranges, relationships, and semantic-search index metadata derived from the current project.

Read the Ontology API guide