Hi everyone
I've been working on a project to integrate Claude AI (by Anthropic) with Genesys Cloud CX to power intelligent Bot Flows - and wanted to share the approach in case it helps others in the community.
---
WHAT WE BUILT
A Node.js REST API that:
- Accepts flow definitions (structured JSON, raw Architect YAML, or natural language)
- Compiles them into valid Genesys Architect YAML using the purecloud-platform-client-v2 SDK
- Deploys and publishes flows programmatically via the Architect API
- Supports all flow types: Inbound Call, Inbound Message, Bot Flow, Outbound Call, In-Queue Call, Secure Call, and Workflow
---
WHERE CLAUDE AI FITS IN
We use Claude as the intelligence layer inside Bot Flows in two ways:
1. NATURAL LANGUAGE → YAML GENERATION
Instead of manually authoring Architect YAML, we describe the desired flow behavior in plain text and Claude generates the full Architect-compatible YAML. This dramatically speeds up flow creation and iteration.
2. RUNTIME NLU + RESPONSE GENERATION
For Bot Flows, we call Claude via the Anthropic API at runtime to handle intent recognition and generate dynamic, context-aware responses - beyond what static utterance matching can do.
---
TECHNICAL APPROACH
The integration works like this:
[Genesys Bot Flow]
→ callData action to external REST endpoint
→ Node.js middleware calls Anthropic API (claude-sonnet-4-6)
→ Response returned as flow variable
→ Bot reads variable via sendResponse action
Key considerations:
• Claude API calls add ~1-2s latency - acceptable for chat, tight for voice
• We cache conversation context in a session store (Redis) keyed by conversationId
• Fallback to static responses if Claude API times out
• PII scrubbing before sending utterances to the external API
---
FLOW TYPES WE'VE AUTOMATED
Inbound Call (IVR with dynamic menu generation)
Bot Flow Voice + Chat (Claude as NLU backend)
Inbound Message (WhatsApp, SMS, Web Chat)
In-Queue Call (dynamic wait messages)
Workflow (backend automation triggers)
---
LESSONS LEARNED
- Architect YAML requires explicit `response` fields and escalation queue refs for Bot Flows - missing these causes intent loops or hard disconnects
- The Genesys `callData` action is the cleanest bridge to external AI services
- OAuth2 Client Credentials grant with Architect > Flow > Edit + Publish roles is all you need
- Test with `POST /flows/compile` (dry-run) before deploying - saves a lot of back-and-forth in the Architect UI
---
Happy to share more details on the architecture, the YAML compiler, or the Claude prompt design for Genesys Bot Flows. Has anyone else experimented with LLM integrations in Architect flows?
Thanks Lucas !
#Architect#Integrations------------------------------
Lucas Diamant
------------------------------