Frequently Asked Questions

Anything you would ask a coding teammate to do. Azent is a general-purpose AI agent (Claude Code or Codex) plugged into your Azure DevOps — within the access you grant through its PAT, what you ask is what it does. Steer it on the spot in an @mention, or pin lasting conventions in a CLAUDE.md or AGENTS.md file.

Most AI coding agents today are built primarily for other DevOps platforms and locked to a single AI provider. Azent is purpose-built for Azure DevOps, lets you bring your own Claude Code or Codex on an API key you control, and runs as a managed SaaS workspace out of the box. If you also need full data residency, the same product can be self-hosted in your own Azure subscription with Azure AI Foundry — same agent, same workflow, just hosted by you. Pick whichever side of that fits.

Azent is available as a managed SaaS service or as a self-hosted deployment in your own Azure subscription. SaaS is the fastest way to start — we host the Azent platform, and you configure your Azure DevOps organization and agent provider. Self-hosted runs the Azent infrastructure in your Azure subscription, giving you more operational control and data sovereignty.

Azure DevOps Services (cloud). Azure DevOps Server (on-prem) is not supported today.

Azent supports Claude through Anthropic and Codex through OpenAI. Depending on configuration, they can be used through direct API access or Azure AI Foundry deployments. Azure AI Foundry is only available when Azent is self-hosted.

Azent does not train its own models on your code. Whether an underlying AI provider may retain or use submitted data depends on the provider, account type, and contract you configure. For stricter control, use Azure AI Foundry or self-hosted deployment.

Azent reads the code needed to complete the requested task. During an agent run, the worker clones the repository, gathers context, and sends relevant context to the configured AI provider. Code is processed for the run; Azent does not claim ownership of it.

Azent uses a bot identity and PAT with permissions needed to read and write code, work items, builds, projects, and wiki content depending on the features you enable. The bot should be scoped to the projects where you want Azent to operate.

Two options, picked per workspace: Api — prompts go to the Anthropic or OpenAI API. Foundry — prompts go to an Azure AI Foundry deployment in your Azure subscription. Code and prompts never leave Azure.

Only repositories and projects allowed by the Azure DevOps permissions and workspace/project configuration. You control where the bot account has access and which projects are enabled.

The Azent platform itself runs in your Azure subscription. AI processing still depends on the AI provider configuration. If you use Azure AI Foundry inside your Azure environment, you can keep the AI path under stronger Azure-side control.

The extension is optional but recommended. It adds trigger buttons and run status visibility directly inside Azure DevOps. Comment-based triggering works independently where configured.

Yes. Azent picks up the standard project-level instruction files used by the underlying AI tools — a CLAUDE.md for Claude or an AGENTS.md for Codex, checked into the repository — so you can preprompt and adjust behavior the same way you would for those tools directly. Use them to capture coding conventions, architecture notes, or anything Azent should know before starting a task.

Very vague requests, high-risk architecture changes without clear direction, production incidents requiring immediate human judgment, and tasks with missing access or unclear acceptance criteria are less suitable.

Yes. It can edit Azure DevOps wiki pages and repository documentation when requested.

Yes. It can create or propose smaller work items from a larger requirement, including suggested titles and descriptions.

SaaS plans are based on included agent hours and concurrency. Larger plans include more monthly agent hours and more concurrent runs.

An agent hour is the time the agent is actively working on a task — the clock starts when the run begins processing and stops when the agent finishes. Queue time before the run starts does not count.

New agent runs are blocked.

SaaS uses the subscription anniversary based on the day you signed up.

Yes. The connected AI model or provider is fully the responsibility of the customer.

The Azent self-hosted license. The customer is responsible for Azure infrastructure costs and the connected AI model or provider.