Provision Foundry Toolbox

Provision Foundry Toolbox

Create a new Azure AI Search index from local text documents, then create an immutable Foundry Toolbox version that exposes the index as azure_ai_search. Always use the Foundry project in AZURE_VOICE_AGENTS_ENDPOINT.

Use the bundled provisioner. It uses the supported azure-ai-projects Toolbox client and discovers the project's Azure AI Search connection. It does not reuse the Foundry IQ URL or connection.

Prerequisites

• Run from the VoiceAgent directory with its Python environment active.
• Install samples/requirements.txt.
• Sign in with az login using an identity that can read the Foundry project, create Search indexes, upload documents, and create Toolbox versions.
• Set AZURE_VOICE_AGENTS_ENDPOINT in samples/.env.
• The Foundry project must contain an Azure AI Search connection.

Workflow

1. Resolve the document folder. Default to samples/sample_foundry_iq_doc.

2. Choose one new lowercase, hyphenated Toolbox name. Add a short unique suffix when the user does not provide a name. Reuse the same name in preflight and creation.

3. Run the read-only preflight:

   python skills/provision-foundry-toolbox/scripts/provision_foundry_toolbox.py --name <name> --dry-run
   

4. Confirm that the plan points to the endpoint project, intended Search connection, correct document count, a new Search index, and a new Toolbox.

5. Create and validate the index and Toolbox version:

   python skills/provision-foundry-toolbox/scripts/provision_foundry_toolbox.py --name <name>
   

6. For another folder, pass --docs-dir <path>. If the project has multiple Search connections, pass --search-connection-name <name>.

7. Return the final two output lines to the user:

   AZURE_VOICE_AGENTS_TOOLBOX_NAME=<toolbox-name>
   AZURE_VOICE_AGENTS_TOOLBOX_VERSION=<immutable-version>
   

8. Add --update-env only when the user asks to replace the Toolbox values in samples/.env.

Behavior and safety

• A normal invocation creates a dedicated Azure AI Search index and a Foundry Toolbox version. Run --dry-run first.
• The provisioner supports UTF-8 Markdown, text, JSON, and HTML files. It chunks each file and creates a lexical index queried with query_type=simple.
• Require a new Toolbox and index name. Do not overwrite or delete existing Toolbox versions, indexes, connections, or documents.
• Never print access tokens or connection credentials.
• On failure, report the failed Search or Toolbox operation. Leave existing .env values unchanged unless --update-env was explicitly supplied and creation completed successfully.