Clara models

What a Clara model is

A Clara model is a complete, named version of how Clara reads a message and decides what to do. Each version is tested as a whole before anyone relies on it.

How versions grow up

There are three simple steps:

1. Preview - an early version we test on. In the product this is the clara-dev channel.
2. Release candidate - a near-final version we trust for last checks. This is clara-staging.
3. Stable - the version customers rely on day to day. This is clara.

You pick a channel when you call Clara. Behind each channel is a pinned version name such as clara-0.1-beta1. Promotion moves the channel pointer to a newer pin after it passes checks.

What each service level does

Clara offers four levels you can adopt one at a time. Every level leaves a record you can defend.

1. Quick answers with receipts - a managed model answers, with a quote and a billable receipt.
2. A fast safe-or-not check - pass, warn, or block on a message, with named reasons.
3. A plain-English briefing - what the customer really meant, risks, and a clean prompt for the next step.
4. The full checked answer - checks, briefing, governed answer, and settlement in one run.

What stays the same across versions

• The record you can defend for each run.
• Test billing during beta (no real charges yet).
• An honest handling story - we only claim what is true today.

What changed

When	What
Beta (clara-0.1-beta1)	First public beta train on the preview channel (clara-dev). Test-mode billing only.

As newer pins ship, this table will grow so you can see progress without reading release notes.