Conversation Workflows Now Available for Telnyx AI Assistants

Conversation Workflows are now available for Telnyx AI Assistants, letting you design multi-step conversations as a directed graph with conditional routing, per-node overrides, and assistant handoff, all on infrastructure built for production voice agents.

What's new

• Workflow nodes: Break complex conversations into focused steps, each with its own label, instructions, and instruction mode (append to or replace the assistant's base prompt).
• LLM and variable-comparison edges: Route between nodes using natural-language conditions for intent-based decisions, or deterministic variable comparisons for account state, channel, duration, and STIR/SHAKEN attestation.
• Per-node model and voice overrides: Switch the LLM or voice for a specific step without changing the assistant-level defaults. Use a faster model for intake, a stronger model for qualification, or a different voice for escalation.
• Per-node tool scoping: Enable or disable individual tools at each step so the model only sees the tools relevant to that conversation stage.
• Assistant routing: Route from one workflow to a different Telnyx AI Assistant when a specialist configuration should take over the call.
• Workflow-aware transcripts: Conversation transcripts show the workflow node associated with each assistant message, connecting real conversations back to the workflow design.
• conversation_flow API: Configure and manage workflows programmatically through the Assistants API with full graph support.

Why it matters

• A single prompt works for open-ended chat. Production voice agents need structure: intake, qualification, booking, escalation, each with different behavior. Workflows turn one assistant into a guided multi-step experience without stitching together multiple bots.
• Each routing hop between vendors adds 30-80ms. Workflows run on Telnyx's private backbone, keeping the entire graph on one system instead of farming each step to a separate platform.
• Deterministic variable comparisons give you predictable routing for compliance-critical decisions like channel type, STIR/SHAKEN attestation, and conversation duration, alongside LLM conditions for intent detection.

Example use cases

• Front desk receptionists that greet callers, identify intent, route to FAQ or transfer, and handle each stage with step-specific instructions.
• Appointment booking flows that guide customers through request, availability, confirmation, and final booking with deterministic gates for required data.
• Escalation workflows that monitor conversation duration and route to a specialist assistant or human agent when troubleshooting exceeds a threshold.
• Multi-assistant support systems where a general assistant routes billing, technical, and sales conversations to specialized assistant configurations.

Getting started

1. In Mission Control, navigate to AI → Assistants → select your assistant → Workflow tab.
2. Enable workflows and add nodes for each major stage of the conversation.
3. Connect nodes with edges and configure the condition type (LLM or variable comparison) for each edge.

For full API reference and configuration details, see the Conversation Workflows documentation.