CoDA

Co-Working Developer Agents

Four AI coding agents. One Databricks App. Three steps to running.

Claude Code, Codex, Gemini CLI, and OpenCode — configured for Unity Catalog, AI Gateway, and Workspace files out of the box.

~/workspace/project

Four Agents, One Terminal

Pick the right model for the job

Different models see different things. Switch agents with a click — they share the same workspace, the same data, the same Databricks context.

Claude Code

Anthropic

databricks-claude-opus-4-6

Deep Databricks skills + useful MCP servers. The most deeply integrated agent.

Codex

OpenAI

databricks-codex

OpenAI's reasoning engine. Excels at multi-step code generation and refactoring.

Gemini CLI

Google

databricks-gemini-2.5-pro

Google's multimodal agent. Vision, long context, and deep reasoning.

OpenCode

Open Source

multi-provider

Open-source, multi-provider. Use any model, any backend, full transparency.

What Ships in the Box

Everything is wired together

Skills, servers, and integrations keep growing. Here's what's configured today.

Databricks Skills

Pipelines, dashboards, Unity Catalog, Lakebase — a growing library.

MCP Servers

DeepWiki, Exa, and more — wired into every agent and growing.

MLflow Tracing

Every agent session auto-traced, queryable via Genie.

Workspace Sync

git commit auto-pushes to your Workspace path.

AI Gateway Routing

One config, any model, full cost tracking.

Terminal Themes

Dracula, Nord, Monokai, and more. Pick your vibe.

Voice & Image Input

Dictate or drag-drop images into the terminal.

Supply Chain Security

All deps SHA-pinned. Weekly CVE audits via GitHub Actions.

Why Databricks Apps?

You bring the code.
Databricks brings the infra.

Running coding agents locally means juggling API keys, model access, and governance. Databricks Apps handles all of that.

Identity & Auth

Your workspace token flows through. No API key juggling. Single-user isolation by default.

AI Gateway

Route agents to any foundation model — Claude, GPT, Gemini — through one gateway. Usage tracked, costs governed.

Data & Governance

Unity Catalog, MLflow, Workspace files — agents have native access to your entire lakehouse.

Databricks Apps gives coding agents what they actually need: identity, models, data, and governance. CoDA just wires it all together.

Need something more specialized?

CoDA agents are general-purpose. Genie Code is bespoke.

The agents in CoDA — Claude Code, Codex, Gemini CLI, OpenCode — are general-purpose coding agents that work across any codebase. They're great for broad software engineering tasks.

But if you need an agent that deeply understands your lakehouse — your table schemas, column lineage, governance policies, pipeline failures — Genie Code is purpose-built for that. It's Databricks' autonomous AI agent for data engineering, data science, and ML work, with native Unity Catalog context that general-purpose agents can't match.

Read the Genie Code announcement

Get Started

Three steps. No Terraform.

1

Fork the template

One click on GitHub. You get the full CoDA setup — agents, skills, MCP servers, themes, and CI.

2

Create a Databricks App

Connect your repo, pick a name. Databricks handles compute, networking, and identity.

3

Set your token, deploy

Add your Databricks token as a secret, hit deploy. Agents start with the app.

Dockerfile Terraform Kubernetes app.yaml
Fork on GitHub