Kentro Chat

apps/kentro-chat is a companion app that lets Kentro's chat product live alongside the existing Python governance engine instead of replacing it.

It includes:

• a React frontend with a ChatGPT-style dark workspace
• a Node/Express backend with POST /api/chat
• an optional backend hook that can call Kentro's existing Python CLI after each reply to generate governance artifacts

Local Run

From the repo root:

cd apps/kentro-chat
npm install
npm run dev

That starts:

• frontend at http://localhost:5173
• backend at http://localhost:5050

The Vite dev server proxies /api/* calls to the Express backend.

Databricks Apps Deploy

To deploy this as a Databricks App, point the app source at:

apps/kentro-chat

This folder is now packaged as a single deployable Node app:

• npm run build builds the React frontend into frontend/dist
• npm run start starts Express
• Express serves both /api/* and the built frontend bundle
• the server automatically listens on DATABRICKS_APP_PORT when running inside Databricks Apps

Required setup notes:

• Do not point Databricks at the broader repo root unless you add separate root-level app packaging.
• If you deploy from a workspace folder, make sure the selected folder is apps/kentro-chat itself.
• If you want the governance hook enabled in Databricks, add the same env vars from backend/.env.example in the Databricks app Environment tab.

Project Layout

apps/kentro-chat/
  app.yaml
  backend/
  frontend/
  package.json
  README.md

Backend Chat API

The scaffold exposes:

POST /api/chat
Content-Type: application/json

Request shape:

{
  "message": "What changed in the latest policy review?",
  "history": [
    { "role": "user", "content": "Summarize our deployment posture." }
  ]
}

Response shape:

{
  "reply": "Scaffolded assistant response...",
  "model": "local-scaffold",
  "governance": {
    "enabled": false,
    "attempted": false
  }
}

Kentro Governance Handoff

By default, chat responses stay local to the Node backend and do not invoke the Python toolkit.

To trigger governance artifacts after each assistant reply, create apps/kentro-chat/backend/.env:

PORT=5050
FRONTEND_ORIGIN=http://localhost:5173
KENTRO_CHAT_MODEL=local-scaffold
KENTRO_ENABLE_GOVERNANCE_HOOK=true
KENTRO_CLI_BIN=tat
KENTRO_CLI_ARGS=
KENTRO_CONFIG_PATH=../../../config.yaml
KENTRO_CONTEXT_FILE=

When enabled, the backend will execute:

tat run prompt --config ../../../config.yaml --prompt "<user message>" --model-output "<assistant reply>"

Useful alternatives:

• If tat is installed in the environment, keep KENTRO_CLI_BIN=tat.
• If the repo is only available as source, set KENTRO_CLI_BIN=python and KENTRO_CLI_ARGS=-m trusted_ai_toolkit.cli.
• If you want a different config, point KENTRO_CONFIG_PATH at another Kentro YAML file.
• If you already have retrieved context in JSON form, point KENTRO_CONTEXT_FILE at that file and it will be forwarded to the CLI.

The chat API still returns a reply even if the governance hook fails. Hook status is included in the JSON response so the UI can surface the result without turning routine chat into a hard failure.

Databricks Job Backend (Option A)

This app can also hand off each question to a Databricks Job instead of running the local CLI hook. In this mode the backend:

1. generates a request_id
2. calls jobs/run-now with question and request_id
3. waits for the Databricks job to finish
4. queries the governance Delta table by request_id
5. returns the final answer plus trust-card summary to the frontend

Enable it by setting these backend env vars:

KENTRO_ENABLE_DATABRICKS_JOB_BACKEND=true
DATABRICKS_HOST=https://<your-workspace-host>
DATABRICKS_TOKEN=<token-with-job-and-sql-access>
KENTRO_DATABRICKS_JOB_ID=<job-id>
KENTRO_SQL_WAREHOUSE_ID=<sql-warehouse-id>
KENTRO_GOVERNANCE_TABLE=wvu.ethanhall.kentroxai_governance_runs
KENTRO_JOB_POLL_INTERVAL_MS=3000
KENTRO_JOB_TIMEOUT_MS=120000

The Databricks job notebook must accept these widgets:

dbutils.widgets.text("question", "")
dbutils.widgets.text("request_id", "")

and persist request_id into the governance Delta table.