Developer Overview
Local development — the whole platform in your own toolset
Developers want to work locally — in their own editor, with their own coding agent, on files they control. Credible is built for exactly that. Semantic models are Malloy code in plain files: edit them in your IDE, version them in Git, review them in pull requests, and publish from your terminal or CI/CD. Nothing about your workflow has to change.
And local development is not a side door. Your local agent runs on the same foundation as every other Credible surface — the same governed models, the same MCP (Model Context Protocol) tools, the same open-source skills. Managed connections mean no credentials ever land on your machine.
Prefer zero setup? You can also build in the app — entirely in your browser, no IDE required. It's the fastest way to get started; local development is best for engineers who want Git-based workflows and direct control over files. Both produce the same governed packages.
The Local Workflow
Local development with Credible looks like development anywhere else:
- Set up your IDE — Install the VS Code Extension (works in VS Code and Cursor, with Claude Code picking up the same workspace config). It equips your agent with managed connections, modeling MCP tools, and modeling skills — the same open tools and skills Malloy Publisher provides.
- Build in files you own — Your agent discovers your data and drafts
.malloymodels right in your workspace. Edit, refactor, and iterate like any other code. - Version and review in Git — Commit models alongside the rest of your codebase; review changes in pull requests.
- Publish from anywhere — From your IDE agent, the CLI, or automatically on merge with CI/CD.
Built on Open Source
The stack you're building on is not a black box. Credible contributed its MCP tools and agent skills to Malloy Publisher, the open-source server for Malloy models: retrieval tools that let an agent look up what your model actually defines, and skills encoding the discipline to use them well — query patterns, gotchas that trip up frontier models, and the rigor that separates a real answer from a plausible one. These are the same skills the in-app agent runs, so how your agent models and analyzes is identical in the app and in your IDE. And because they're open, they're curated with the world's data experts — read them, fork them, or extend them with your organization's institutional knowledge. Read the full story in We Open Sourced the Thing Everyone Else Is Selling.
Set Up Your Tools
VS Code Extension
Build models in VS Code, Cursor, or Claude Code — managed connections, modeling MCP tools, and skills, no credentials on your machine
Connect your Coding Agent
Let any coding agent analyze data with your published models — the same governed access as workspace chat
Embedded Data
Ship CSV and Parquet files inside your packages for standalone, versioned models
CLI
Everything the Credible App manages, scriptable from your terminal