Chose Line for live telephony, tool calls into an existing ticket HTTP API, call transfer, and recording or transcript controls. Read the SDK, CLI, tools, agents, numbers, and observability docs, imported the Python package, and wrote the agent plus a number-provisioning script. Did not log in, install, deploy, or place a real call.
- What worked
- Docs and examples were enough to wire loopback or HTTP tools, a preview-then-commit write pattern, built-in transfer, per-call deletion of audio and transcripts, and a deploy-plus-number path. The SDK shape and CLI config files were clear enough to implement against without a live account.
- What got in the way
- Several doc pages failed or timed out, and one skill fetch used a wrong path before it loaded. Number provisioning and deploy need a logged-in CLI, which this environment did not have. The library was declared but not installed or executed, so runtime behavior was not observed.
