ONYX AI · BUILT AROUND YOUR DATA

Your database
has answers.
Now it can help
you find them.

Ask a question. Investigate a pattern. Build an agent. Train a model. Put the result in your application. Onyx connects AI to the data and tools you already use to build.

Start with Onyx Explore the platform
ONE CONNECTED PLATFORM

Database → AI → Application

See how it works

The idea behind Onyx AI

Start with the data.
Go as far as you want.

Most product questions begin in your data: a customer record, a transaction, a pattern across thousands of rows. Onyx puts AI beside the database, query tools, development workspace, and application interfaces.

Start with a plain-language answer. When the question grows, let an agent investigate. When the answer needs to become a feature, call AI through the SDK or publish an agent or model for your app.

Explore the platform

Six ways to turn data into action

Choose the right depth for the job, from a quick database question to a shipped application feature.

01 / Understand

AI Chat

Ask about a selected database in plain language. The assistant can use its schema, saved queries, and query tools to help explain and explore your data.

A fast answer when you need to know what your data says.

Explore AI Chat
02 / Investigate

Agentic Research

Give an agent a larger question. It can examine records in batches, track its progress, and return an answer with supporting analysis and files.

Grant actions when you want it to create scripts, visualizations, or dashboards.

Explore Agentic Research
03 / Automate

Agent Builder

Design an agent as a workflow of connected steps. Choose prompts, models, tools, and approval paths; test it in chat before publishing.

Call the published agent by model identifier from your application.

Explore Agent Builder
04 / Predict

Model Builder

Use a saved database query as training data. Map input features and targets, train a neural network, inspect runs, and publish a model.

Send application inputs to its prediction API.

Explore Model Builder
05 / Build

Workspace AI

Work with an AI coding agent beside your source code and live preview. Ask it to implement features, inspect changes, and run checks in your workspace.

Your app, data connection, and development tools stay in one place.

Explore Workspaces
06 / Integrate

SDKs & AI API

Call chat from your code with database credentials. Use the SDK for short answers or the chat completions API for messages, streaming, and agents.

Bring intelligence into a real product flow.

Browse the SDKs

01 / Understand

Ask the database a real question.

You can open AI Chat for a selected database and ask in ordinary language. Onyx can use the database schema and query tools to help find an answer, draft a script, or explain what is available. Saved queries and documents can provide additional context.

  • 01 Ask about records, trends, or the shape of your data.
  • 02 Inspect generated queries and results.
  • 03 Keep the conversation as you refine the question.
See the AI Chat guide

Working in the Query Editor? Its AI Query Assistant can draft and revise queries there, too.

Onyx AI ChatDatabase: support
YOU ASK

Which open support issues are affecting our largest accounts?

ONYX INVESTIGATES

Look across your account and issue tables, inspect the relevant records, and explain the finding in context.

Schema contextLive queriesFollow-up questions
Illustrative question · Results depend on your database and permissions

02 / Investigate

Some questions deserve more than one response.

Agentic Research works through a longer investigation. It can page through large datasets, show what it is checking, and preserve task history when you leave the page. The answer includes evidence and available supporting files.

When you grant the relevant actions, research can also create saved scripts, visualizations, dashboards, and documents. Its permissions are checked against your role and the selected database.

How Agentic Research works
01Ask

“What changed in customer activity this month?”

02Investigate

Query in batches, analyze patterns, show progress.

03Deliver

Answer, evidence, and supporting files.

AI in your code

An AI answer can become a database query.

Use the same Onyx client to ask a question and query structured records. The SDK makes it straightforward to compose the two in an application flow.

TypeScript · @onyx.dev/onyx-database
// Ask AI for a value, then use it in a query.
import { onyx, eq } from "@onyx.dev/onyx-database";

const db = onyx.init();
const name = (await db.chat(
  "Who was the first U.S. president? Reply with only the full name."
)).trim();

const users = await db.from("Users")
  .where(eq("name", name))
  .list();
AI supplies a value. Onyx queries your records.
01

Ask naturally

db.chat() returns the answer text from the configured AI model.

02

Use the result

Await the answer before passing it to where(eq(...)). Your application can validate the value before querying.

03

Go further

Use chat completions for multi-message conversations and streaming, or call a published agent with its model identifier.

Read the TypeScript SDK referenceExplore the AI API

The code is illustrative: your schema needs a Users table with a name field. Exact-match filters should use a validated model response.

From answers to capabilities

Build intelligence your application can call

Onyx gives you two routes from an idea to a reusable AI capability.

Agent Builder

Design the steps. Publish the agent.

Connect workflow nodes, choose tools and models, and test how the agent handles real questions and approvals. Publish a model identifier, then call the agent through chat completions.

DesignTestPublish
Build an agent

Model Builder

Train on queries. Publish predictions.

Start from a saved Onyx query, map input features and targets, configure and train a neural network, then publish a version for prediction requests from your application.

Map dataTrainPredict
Build a model

Workspace AI control plane

And when you are ready to ship, keep building in Onyx.

Workspaces bring the AI coding agent, source files, live preview, tests, and environment controls into one development surface. Ask it to implement a feature, inspect its diff, run checks, and connect your application to the database and AI capabilities above. You keep the code and review the path to release.

Explore Onyx Workspaces

Good questions

A few things to know.

Is AI Chat the same as Agentic Research?

AI Chat is useful for focused questions about a selected database and for help with queries or scripts. Agentic Research is a separate, longer running workspace for investigations that can page through large datasets, show progress, keep task history, and return supporting analysis. Actions such as saving resources require the permissions you grant.

Can I use Onyx AI inside my application?

Yes. The Onyx SDK supports db.chat() and AI model calls. The OpenAI compatible chat completions endpoint accepts a database context and can invoke a published Agent Builder workflow by its model identifier. Published Model Builder models are available through a separate prediction API.

Does my application need to export data for AI?

Database chat and queries can use your selected Onyx database directly. Model Builder trains from a saved Onyx query. You can also add documents for supporting context. The access and tools available depend on your credentials, model, and permissions.

Your next question could be the start

Put your data to work.

Start with a question. Build the intelligence your product needs.

Get started with Onyx Explore documentation