Build with AI coding agents

Build and test Voximplant voice apps using coding agents
View as Markdown

For the complete documentation index, see llms.txt.

Voximplant AI Agent Skills give coding agents current, task-specific instructions for writing VoxEngine scenarios and managing Voximplant resources. Describe the call experience you want, review the proposed changes, and let the coding agent build, deploy, and test the first version with you.

Each coding agent has its own installation process, permission model, and testing workflow. See the Voximplant AI Agent Skills repository for Cursor and Codex installation instructions. More detailed guides are coming soon.

How the skills work together

Most complete Voximplant workflows use both skills:

  • voximplant-voxengine-dev writes and reviews the VoxEngine scenario, verifies platform APIs, and fixes code-level call behavior.
  • voximplant-management-api creates applications, scenarios, routing rules, secrets, and test users. It also retrieves call logs and recordings for debugging.

The skills use progressive disclosure: your coding agent loads the detailed instructions only when your task needs them. This keeps the working context focused while grounding generated code in Voximplant documentation instead of relying on model training data.

What the coding agent can do

For a first project, the agent can guide you through this loop:

1

Describe the call

Explain the inbound or outbound experience you want in plain language.

2

Review the plan

Confirm the target account, application, resources, and credentials before the agent changes the platform.

3

Build and deploy

The agent writes the VoxEngine scenario, configures Voximplant resources, and uploads provider credentials to Voximplant Secrets.

4

Place a test call

Test without buying a phone number by using a Voximplant application user and the browser phone.

5

Debug from evidence

If the call fails, the agent can retrieve the latest call logs, identify the failure, update the scenario, and repeat the test.

Generated code and platform changes still require review. Keep service account files and provider credentials out of source control, and approve account-changing operations only after checking the proposed resources.

Build your first voice agent with Claude Code