Prompt Versioning and Source Control

Bundled into the existing Operationalizing Machine Learning and Generative AI Solutions premium course — no separate purchase.

14-day money-back guarantee — no questions asked.

Included in this chapter:

Aspect	Prompt agent	Hosted agent
What you author	Declarative config: instructions, a model, and tools	Your own code using any agent framework
How it's versioned in source control	Config files committed to Git	Code in Git, built and pushed as a container image to Azure Container Registry
Compute to run and pay for	None; Foundry runs it with no compute to manage	Container compute you pay for; deprovisioned after 15 minutes idle
Runtime identity	Managed by Foundry	Dedicated Microsoft Entra agent identity minted per agent at deploy
Versioning and rollback	Built-in version snapshots; promote through CI/CD	Each deploy is a new immutable version; previous versions kept for rollback
Choose when	Agent needs only a model, instructions, and built-in tools	Agent must run custom orchestration or framework code

Cheat sheet

A prompt's system message and instructions define agent behavior and grounding
Store prompts as code assets in a Git repository
A repo structure separates prompt versions and environments
Compare prompt versions with Foundry evaluations on a fixed dataset
Prompt flow is retiring; build new prompt and agent orchestration on Agent Framework
Define agents and prompts as configuration and promote them through CI/CD
Choose a prompt agent or a hosted agent to operationalize an agent
Hosted-agent deployment lifecycle: versioned deploys, auto identity, tracing, and scale-to-zero

Unlock with Premium — includes all practice exams and the complete study guide.

Cheat sheet