quickstart

name	quickstart
description	Set up the Databricks chatbot app for local development and deployment. Use when: (1) First time setup, (2) User says 'quickstart', 'set up', 'authenticate', or 'configure databricks', (3) No .env file exists, (4) User says 'enable feedback', 'feedback widget', or 'MLFLOW_EXPERIMENT_ID'.

Quickstart: Databricks Chatbot App

Run the interactive setup script from the e2e-chatbot-app-next/ directory:

./scripts/quickstart.sh

What It Configures

The script walks through the following steps interactively:

1. Prerequisites — installs jq, nvm, Node 20, and the Databricks CLI if missing
2. Databricks auth — lets you select an existing profile or create a new one via databricks auth login
3. Serving endpoint — prompts for your agent endpoint name (Agent Bricks or custom agent); validates it exists
4. Feedback widget — if the endpoint has a linked MLflow experiment, offers to enable thumbs up/down feedback (default: yes). This automatically:
   • Sets MLFLOW_EXPERIMENT_ID in .env
   • Uncomments the experiment resource in databricks.yml
   • Uncomments MLFLOW_EXPERIMENT_ID in app.yaml
5. App/bundle name — optionally customize the app name (default: db-chatbot-dev-<username>, max 30 chars)
6. Database — optionally enable persistent chat history via a Lakebase instance (~5-10 min, costs apply)
7. Deploy — runs databricks bundle deploy and starts the app

Enabling Feedback After Initial Setup

If you skipped feedback during quickstart, or ran the script before feedback support was added, enable it manually:

1. Find the experiment name for your endpoint:

npx tsx scripts/get-experiment-id.ts --endpoint <your-serving-endpoint-name>

For Agent Bricks (Knowledge Assistant / Multi-Agent Supervisor):

npx tsx scripts/get-experiment-id.ts --agent-brick <agent-brick-name>

2. Set MLFLOW_EXPERIMENT_ID in .env (for local dev):

MLFLOW_EXPERIMENT_ID=<experiment-name-from-step-1>

3. Configure databricks.yml — uncomment and fill in the experiment resource:

        - name: experiment
          description: "MLflow experiment for collecting user feedback"
          experiment:
            name: "<experiment-name-from-step-1>"
            permission: CAN_EDIT

4. Configure app.yaml — uncomment the env var:

      - name: MLFLOW_EXPERIMENT_ID
        valueFrom: experiment

5. Redeploy:

databricks bundle deploy
databricks bundle run databricks_chatbot

Troubleshooting

Issue	Solution
Feedback widget not showing after setup	Restart dev server; env vars are read at startup
get-experiment-id.ts fails with auth error	Run databricks auth login first
No experiment found for endpoint	Only custom agents and Agent Bricks endpoints have linked experiments; Foundation Model endpoints do not support feedback
Feedback submission returns 403	App service principal is missing CAN_EDIT on the experiment — check permission: CAN_EDIT in databricks.yml
"Instance name is not unique" on deploy	Run ./scripts/cleanup-database.sh to remove the old database instance