Support Hosted Voice Agents in azd
- Dominant language
- Go
- Stars
- 569
- Forks
- 364
- Avg merge
- 2d 19h
- Merged PRs (30d)
- 136
Description
## Goal
Add first-class azd support for Hosted Voice Agents while preserving the existing Hosted Agent and `invocations_ws` experience.
## User experience
### Initialize from a sample manifest
```powershell
azd ai agent init -m
azd provision
azd deploy
```
### Add existing source to an azd project
```powershell
cd
azd ai agent init
azd provision
azd deploy
azd deploy
```
## Requirements
- Generate/deploy a Hosted Agent target using Voice Bridge Protocol 1.0 over `invocations_ws/1.0.0`.
- Generate/deploy a Voice wrapper with `model_type=hosted_agent` and a pinned target version.
- Reuse normal project/model selection, package, publish, provision, deploy, show, doctor, and delete behavior.
- Support standard Agent Manifest initialization from an empty folder and local-manifest detection in an existing project.
- Validate target project, status, protocol, and Voice Bridge metadata before wrapper deployment.
- Preserve unrelated services when adding Hosted Voice to an existing azd project.
## Sample
Current source:
- Repository: `https://speedme.visualstudio.com/TextToSpeech/_git/voice-first-agent-dev`
- Path: `samples/voice-hosted-agent/voice-hosted-agent-dotnet`
A public GitHub Agent Manifest URL remains a follow-up for sample publication.
## Validation
- Unit, lint, schema, and cross-platform build coverage.
- Empty-folder manifest initialization.
- Existing-source/project initialization.
- West US 2 target/wrapper deployment.
- Typed `/help` routing through the wrapper.
- Real audio STT -> target -> TTS.
- Portal and sample Dashboard validation.
Contributor guide
Research direction
Start with the azd ai agent init, provision, and deploy entry points and compare the existing Hosted Agent and invocations_ws behavior. Use samples/voice-hosted-agent/voice-hosted-agent-dotnet as the integration reference. Done includes manifest initialization, target and wrapper validation, deployment, preserved services, and the listed unit, schema, build, audio, and Dashboard checks.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- azure, go
- Domain
- ai, cli, cloud
- Issue type
- Feature
- Difficulty
- 5/5
- Estimated time
- Over a week
- Activity status
- Active
- Clarity
- Mostly clear
- Newbie friendliness
- 35/100