Admin access requires a @netrefer.com email address. All admin UI is under the Admin tab in the platform.

---

Environment variables

Variable	Required	Description
DATABRICKS_SERVER_HOSTNAME	Yes	Databricks SQL warehouse hostname
DATABRICKS_HTTP_PATH	Yes	Databricks warehouse HTTP path
DATABRICKS_TOKEN	Yes	Databricks personal access token
B2C_CLIENT_ID	Yes	Azure B2C application (client) ID
B2C_TENANT_HOST	Yes	B2C login hostname (e.g. example.b2clogin.com)
B2C_TENANT_NAME	Yes	B2C tenant name
B2C_POLICY	Yes	B2C user flow policy name
B2C_REDIRECT_URI	Yes	OAuth redirect URI for the SPA
ANTHROPIC_API_KEY	No	Can be set here or via the AI settings UI
CONFIG_ENCRYPTION_KEY	No	Fernet key for encrypting stored secrets at rest. Generate: python -c "from cryptography.fernet import Fernet; print(Fernet.generate_key().decode())"
SQL_CONNECTION_STRING	No	SQL Server connection string. Falls back to local JSON file if unset.
CORS_ALLOW_ORIGINS	No	Comma-separated allowed origins. Default: http://localhost:6274
PORT	No	Service port. Default: 8000

---

Table configuration

Adding tables

Tables in the Databricks gold_data schema with the ds_ prefix load automatically. To add tables from other schemas:

1. Go to Admin > Tables > Explore layers to browse available schemas.
2. Find the table you want to add.
3. Click Add to platform.

Column configuration

For each table you can configure:

• Enabled columns — control which columns the AI uses. Disable PII or irrelevant columns.
• Column descriptions — plain-English descriptions. Click Auto-describe to generate with AI, then review and save.
• Aliases — map a column name to a business-friendly alias (e.g. subscriber_guid → client).
• Cast overrides — override the data type shown to the AI.

Table descriptions

Add a plain-English description for each table. Click Auto-describe table to generate one, or write it manually.

Schema refresh

The schema is cached for 30 minutes. To force a refresh after adding a new column to Databricks, go to Admin > Schema > Refresh.

---

Glossary

The glossary maps business terms to SQL terms. The AI uses this when generating queries.

Examples:

• revenue → sum(deposit_amount - withdrawal_amount)
• active player → player who made a deposit or withdrawal in the last 30 days
• FTD → first time depositor

Go to Admin > Glossary to add, edit, or delete terms. Click Recommend to have the AI suggest terms based on your table schemas.

---

Query examples

Verified examples show the AI exactly how to write certain queries (few-shot prompting).

To add an example: ask a question in chat, get a correct result, then click Save as example.

To manage examples: go to Admin > Query examples. Click Compact to merge similar examples and reduce prompt size.

---

AI model settings

Go to Admin > AI settings to configure:

• Provider — Anthropic (default) or OpenAI
• Model — which model to use for NL→SQL generation
• API key — enter here or set ANTHROPIC_API_KEY / OPENAI_API_KEY env var

Changes take effect immediately — no restart required.

---

Tool permissions

Go to Admin > Tool permissions to enable or disable individual tools per integration. For example, you can enable Confluence search but disable the Atlassian tool that creates Jira tickets.

---

Failed queries

The platform logs queries the user flagged as incorrect or that returned errors. Go to Admin > Failed queries to:

• Review — see the question, the SQL generated, and the error or user feedback.
• Correct — write the correct SQL. This automatically adds it as a query example.
• Dismiss — mark as not actionable.
• Generate training questions — create test questions from current table schemas.

---

Integration setup

Atlassian / Confluence (OAuth)

1. Go to Admin > Integrations > Atlassian.
2. Enter the Atlassian OAuth app credentials (Client ID, Client Secret).
3. Set the Confluence base URL.
4. Click Start OAuth — you will be redirected to Atlassian to authorise.

Token refresh is handled automatically. If the token expires, re-run the OAuth flow.

HubSpot

Two modes: Shared token (one HubSpot connection for all users) or Per-user token.

1. Go to Admin > Integrations > HubSpot.
2. Enter the HubSpot OAuth app credentials.
3. Choose shared or per-user mode.
4. Click Start OAuth to connect.

Azure Cost Tools

1. Go to Admin > Integrations > Azure MCP.
2. Enter: Azure tenant ID, Client ID, Client secret, MCP server URL (typically http://localhost:3001), MCP app ID.
3. Save. Users can then connect their own Azure accounts via OAuth.

Coralogix

1. Go to Admin > Integrations > Coralogix.
2. Enter the Coralogix API key and endpoint URL.
3. Save. Changes take effect immediately.

n8n workflows

1. Go to Admin > n8n.
2. Enter the n8n API URL and API key.
3. Click Sync workflows to import the workflow list.
4. Use Get workflow metadata to add display names and descriptions.

---

Config persistence

By default, all configuration is stored in a local JSON file (data/config.json). For shared/persistent config, set SQL_CONNECTION_STRING to a pyodbc-compatible connection string:

Driver={ODBC Driver 18 for SQL Server};Server=myserver.database.windows.net;Database=mcp_config;Uid=myuser;Pwd=mypass;Encrypt=yes;

If CONFIG_ENCRYPTION_KEY is set, all secrets stored in the config (API keys, OAuth tokens) are encrypted at rest using Fernet (AES-128-CBC). Set it in production.