Websites had their WordPress moment. SaaS is having its BuildWithHQ moment. See the new BuildWithHQ homepage →
Home/Templates/AI configurations
Industry-specific AI configurations

AI tuned to your industry, tested before it goes live

An industry-specific AI configuration is everything that sits around the model for one industry: instructions, tone, vocabulary, worked examples, permitted actions, guardrails, and compliance language. It is versioned like any other template, and it cannot be published until it passes its own tests. In the product these are called harness packs.

What is inside one

Configuration, not code

See harness packs in the console

System prompts

The ground rules the assistant operates under, with variables for the application, tenant, and user.

Personas

Tone and style rules, so the assistant in a legal product does not sound like the one in a field service product.

Glossary sets

The industry's own vocabulary, so the assistant uses the customer's terms rather than generic ones.

Few-shot exemplars

Worked examples that shape output format and reasoning on the tasks that matter in that vertical.

Action sets

Which actions the assistant may propose in this industry, and which it may not.

Guardrails and notices

Refusals, escalations, and compliance language appropriate to the domain.

Layering

Four levels, resolved per component

Packs stack rather than replace. Each layer overrides only the components it defines, so a tenant can adjust one glossary without inheriting the responsibility for an entire configuration.

  • Platform base — the ground rules every application inherits.
  • Industry pack — vertical vocabulary, personas, and guardrails.
  • Builder layer — your product's own adjustments across all your customers.
  • Tenant layer — one customer's terminology and preferences.

Resolution picks the highest-precedence active definition for each component: tenant, then builder, then industry, then platform base. It is the same override philosophy used for page forks, applied to AI configuration.

Evaluations as a publish gate

A pack can carry an evaluation set flagged as a publish gate. When it does, publishing a version is refused until a passing evaluation run exists for that exact version. Not a score to review afterwards — a gate that blocks the release.

Behavior, never access

Nothing in a harness pack can widen a permission envelope. Packs shape how the assistant behaves; the security boundary is enforced beneath this layer and no payload can reach it.

Common questions

Questions about this

Who writes industry packs?

Platform-authored packs cover the base. Vertical experts can author and publish their own, and partner packs can carry a certification status so buyers know what has been reviewed.

Can a pack make the assistant see more data?

No. Packs configure behavior only. Permission enforcement sits below this layer and is not reachable from a pack payload.

What if a pack's evaluations fail?

The version does not publish. That is the point of a gate rather than a report.

Get started

Build it, then prove what it did

Create an account, explore the full builder console, and launch your first application on a custom domain. Nothing is charged while your application is still in development.

Start building today

No credit card required. The console preview, the documentation, and the architecture reference are all readable before you sign up.

Get started free Read the docs