Talk to an Expert

Built on Databricks

Halo runs in your Databricks account

Talk to an Expert

The Challenge

Modern IT organizations in retail and CPG have spent years consolidating systems and building the fundamentals for becoming data-driven companies. On their own, those fundamentals don’t produce business value: that requires additional investment in the user-facing applications built on top of them, including Crisp.

These solutions have traditionally been delivered as external applications, hosted outside the customer’s own environment. That model has real advantages: short time to value, and a simple, predictable total cost of ownership. It also comes with real drawbacks for enterprise IT: an extended data governance perimeter, complex access control across systems, and contractual hurdles related to data rights. Although Crisp has always been able to integrate with a customer’s data systems, significant portions of the Crisp technology stack have still run externally, carrying those same drawbacks.

The Solution

Crisp Halo deployed within your own infrastructure

The Crisp Halo platform can now be fully deployed inside customers’ Google Cloud, Databricks and Snowflake environments.

Uniform access control

User management and authorization is done using the existing capabilities (such as Unity Catalog) within the target environment. The retail-specific semantic layer can easily be integrated with row level security based on client needs.

Zero data exports or sharing

Ingestion happens within the client platform and no proprietary data ever leaves the client’s environment. Only non-proprietary log and telemetry data is shared and monitored by central Crisp systems.

Easier data integration

Halo Agents and Applications can connect to any authorized dataset within the data warehouse without data pipelines.

Seamless user experience

Users can still enjoy the rich Halo user experience.

Flexible resource allocation

Halo can leverage existing or new compute and inference commitments.

Full support and maintenance

Crisp manages and supports the entire platform even without being able to see any of the data under management.

Architecture

How Halo is deployed in Databricks

Built on your Databricks investment

Crisp Halo uses the Databricks capabilities you already operate rather than introducing a separate platform.

Databricks capability

How Crisp Halo uses it

Unity Catalog

Crisp Halo’s files live in managed volumes and its data in Delta tables, in a catalog you choose. Each Crisp Halo service gets only the grants it needs.

Databricks Apps

Hosts the Crisp Halo web app and agents, background tasks and schedules, and the semantic query service. Users sign in through the Apps proxy with workspace SSO.

Lakebase Postgres

One managed Postgres database holds conversations, projects, agent memory with vector search, the live response stream, and the task queue.

Mosaic AI Model Serving & AI Gateway

Runs every model call, with the models and providers you choose, under your rate limits, guardrails, and usage tracking. Usage appears in your system tables.

Databricks SQL

Runs semantic queries for chat, apps, and boards on a SQL warehouse you choose.

UC Metric Views

Rosetta models ship with each release and compile to Databricks SQL. Publishing them as UC Metric Views for your other tools is on the roadmap.

Lakeflow Jobs & Lakeflow Pipelines

Lakeflow Jobs load retail data into Delta tables. Scheduled reports and background agent tasks run in the Crisp Halo apps.

Unity Catalog Volumes

Stores project files, uploads, and generated reports and presentations.

MLflow

On the roadmap. Today token usage is recorded for every run and in your system tables, and app logs are available in Databricks Apps.

Identity & system tables

Signs users in through your workspace SSO; workspace groups decide who can use and administer Crisp Halo. Usage is reported in your system tables.

What gets installed in your workspace

Each release is one Databricks Asset Bundle that creates or upgrades these resources in your workspace.

Resource

What it does

Databricks Apps

The Crisp Halo web app and agents, a worker for schedules and background tasks, and the semantic query service.

Lakebase database

Application state, agent memory with vector search, the live response stream, and the task queue.

Unity Catalog schema and volumes

Project files, uploads, generated reports, and retail data tables.

SQL warehouse

Your existing warehouse, or one the installer creates.

Model Serving and AI Gateway

The models you choose, from any provider you enable.

Secret scope

Keys that encrypt stored credentials.

Workspace groups

Who can use Crisp Halo, and who administers it.

Deployment and operations

Installed in minutes, kept current automatically.

Customers can choose between weekly and daily release schedules. Regular releases have zero down-time.

The installation of Crisp Halo is fully automated and all resources will be fully provisioned within minutes. Once the system is deployed, it will be kept up to date by listening to new versions being published by the central Crisp package systems. System operational logs (no proprietary data) will be replicated to Crisp central systems to detect operational anomalies and uptime issues. Built-in AI Agents help debug issues automatically and share sanitized bug reports with central Crisp systems.

Support and maintenance

Support maintenance contracts are similar to existing Crisp support contracts; we also accept contractor status for named personnel with custom security training, NDAs and system access through customer-provisioned laptops / endpoints, as needed.

See Halo run in your cloud

Speak with one of our retail experts to plan a deployment in your Databricks, Snowflake, or Google Cloud environment.

Talk to an expert