CRAIM already contains an AI calls module. This should be documented as a real product area, not as a future placeholder.
Current product shape
The backend exposes endpoints for:
- creating call tasks
- launching call tasks immediately
- reading call sessions
- saving call sessions
- saving call analytics
Architectural principle
AI calling should be treated as a dedicated execution surface inside the broader AI sales operating model.
That means:
- voice runtime is specialized
- CRM and policy ownership stay in backend control
- post-call actions should flow back into the general sales loop
Telephony contours
The internal architecture already distinguishes between:
- built-in telephony paths
- client-owned telephony providers
Why that is correct
Telephony has different latency, transport, and provider behavior than text workflows. It should not be forced into the same execution loop as normal message drafting.
Product guidance
User-facing docs for AI calls should explain:
- what a call task is
- how a session relates to a lead or deal
- how provider selection works
- where transcripts and analytics end up
- what still requires human review or policy approval
Operating rule
Treat AI calls as one execution mode of the AI sales system, not as a disconnected product.