Build & Publish

The Credible App (https://<your-org>.app.credibledata.com) lets you build semantic models without leaving your browser and without any local setup. Everything happens through a natural-language agent: you describe what you want, and the agent connects your data, drafts the Malloy model, previews results, and publishes — guiding you the whole way.

The in-app experience is designed for getting productive fast and for business users who don't want to manage an IDE or CLI coding agents. It handles the vast majority of modeling workflows with a fraction of the complexity of a local setup.

Prerequisites

• A Credible account in a Credible organization — create a new organization or get added to an existing one; every account comes with a personal workspace to start in
• A data source — either a connection your admin has configured, or one you add through the agent's private connection form. No data handy? The form includes a sample dataset to explore with

That's it. There is nothing to install.

What the In-App Agent Does

The in-app user experience is fully agentic and natural-language first, and the agent can do everything you do in Credible — build semantic models, analyze data, build data apps and reports, and manage your environments — all from a single chat. Its behavior comes from Credible's open-source agent skills, the same skills that drive local development — so how it models, analyzes, and publishes is consistent everywhere.

This page covers the core build loop — models, data apps, and publishing. Within your environment, the agent can:

• Connect your data — set up an environment and connect your warehouse through a private in-chat form; Credible indexes the schema so the agent can find tables by meaning
• Explore your data — discover tables, columns, and relationships, and suggest how to model them
• Build the model — author fully documented Malloy sources, joins, dimensions, and measures in a draft package
• Preview as it goes — run queries against the draft so you can validate results before publishing
• Build data apps and reports — generate interactive dashboards and data apps that ship alongside the model (see Building a Data App)
• Publish and share — deploy the finished package so it's available everywhere in Credible, and open a sharing panel so you choose who gets access

Building a Model

1. Open your workspace. The agent introduces itself and asks about your data.
2. Connect your data. If nothing is connected yet, the agent sets up an environment and opens a private connection form in the chat — pick your warehouse type, enter credentials, and choose which tables to index. Credentials go straight to Credible; the agent never sees them. No data handy? The form offers a sample dataset.
3. Describe what you want to model. Tell the agent about the questions you want to answer (e.g., "model our orders and customers so I can analyze revenue by region").
4. Confirm the agent's proposals. The agent researches before it asks: it explores your schema, then proposes — backed by real data — which tables to include, how sources join, and which dimensions and measures to define. Confirm or adjust each proposal in plain language.
5. The agent builds the draft. It writes fully documented Malloy files — every field defined, documented, and indexed for discovery — into a draft package you can open from your workspace at any time. It previews queries as it goes, so you see real results at each step.
6. Review and publish. The agent presents the model's structure and assumptions for a final review, and can optionally propose access controls and discovery curation. When you're happy, ask it to publish. See Publishing below.

Building a Data App

Once you have a model in your draft package, ask the agent to build a data app on top of it — an interactive dashboard or application that ships with the package:

1. Describe the app. Tell the agent what you want to see — the charts, filters, and layout (e.g., "build a revenue dashboard with monthly trends, filterable by region").
2. Iterate. The agent generates the app into the draft package as plain web files in its public/ directory. Preview it, then ask for changes in plain language — new charts, different breakdowns, styling.
3. Publish together. When the package is published, the data app ships with it — governed by the model's access rules, versioned with the model it draws from, and listed automatically in the Data Apps section of every workspace the package is added to. Apps can even hand questions back to the in-app agent, so viewers can go from a dashboard number to "why?" in one click.

See Build Data Apps for how data apps work and how they're used.

Publishing

Publishing happens right from the chat — ask the agent to publish your draft, and it packages it, deploys it to your environment, and confirms the published version. The agent only publishes when you ask; it never publishes on its own.

After publishing, the Credible AI Analytics Engine indexes your model — usually a minute or two — and it's ready for consumption. You can then:

• Keep going in the same chat — ask questions of the published model, or ask the agent to create your first report
• Chat with your model and build reports and data apps in any workspace the package is added to
• Connect your agent — chat with the model from Claude, ChatGPT, Gemini, or any MCP client
• Share it — the agent opens a private sharing panel where you choose which teammates and groups get access (see Permissions)

For versioning details — pinning a version as "latest", testing before promoting, and version history — see Publishing.

Next Steps