Encyclopedia Evalica / Deployment / Prompt

Prompt illustration

Prompt

/prahmpt/A versioned, first-class object in Braintrust that can be managed, compared, and deployed. Prompts being versioned enables fast iteration and safe rollbacks. (noun)

Why it matters

A prompt change can alter your system's behavior as dramatically as a code change, but prompts are often edited informally and deployed without evals. This is risky because small wording changes can cause large shifts in output quality, tone, or safety behavior, and without versioning you have no way to roll back or compare. Treating prompts as versioned, first-class objects means every change is tracked, every version can be compared against previous ones in an experiment, and rolling back is a configuration change rather than an emergency code deploy. This also enables a faster iteration workflow. Instead of editing a prompt in code, committing, deploying, and waiting to see results, you can update a prompt, run it against your eval suite, verify the change helped, and promote it to production, all without touching application code. Teams that manage prompts this way iterate significantly faster because prompt changes are decoupled from the software release cycle.

We evaled the new prompt version then deployed once we were confident in our changes.

Related Deployment terms

From the docs

Get started with Evals

Braintrust is the AI observability and eval platform for production AI. By connecting evals and observability in one workflow, teams at Notion, Stripe, Zapier, Vercel, and Ramp ship quality AI products at scale.

Start building