hume-evi-langgraph

What This Skill Does

The hume-evi-langgraph skill provides a robust architectural blueprint for building sophisticated voice-based AI agents. It bridges Hume AI’s EVI (Empathic Voice Interface) with LangGraph's state management capabilities. By leveraging an interrupt/resume pattern, it allows developers to handle the asynchronous lifecycle of a phone call—starting from the initial Twilio trigger, through the real-time voice conversation, and finally to the post-call analysis of sentiment and emotion.

Installation

You can install this skill directly via the OpenClaw CLI using the following command: clawhub install openclaw/skills/skills/brandonwadepackard-cell/hume-evi-langgraph

Use Cases

This skill is ideal for:

• Customer support agents that require sentiment-aware post-call summaries.
• AI-driven coaching assistants that analyze user frustration or satisfaction levels during live calls.
• Dynamic appointment schedulers that need to maintain stateful context across multiple interaction phases.
• High-fidelity empathetic bots that adjust their tone based on user emotional output.

Example Prompts

1. "Initialize a new Hume EVI session using the KORA voice and set the system prompt to be a empathetic medical triage assistant."
2. "Update my LangGraph workflow to fetch the last call's transcript and extract the top 5 dominant emotions from the user's side of the conversation."
3. "Setup a webhook listener that triggers the analysis state in my graph once the Hume chat session status changes to ended."

Tips & Limitations

• Temperature Control: Always cap the temperature at 0.6. Hume's default settings can be overly eager; this restriction ensures a more professional, grounded agent persona.
• Endpoint Handling: Do not query the /chats/{id}/events endpoint directly, as it often returns a 404. Instead, resolve the chat_group_id first and query the chat_groups events endpoint.
• TwiML: Use & for XML escaping within the TwiML redirect strings to ensure the URL parameters are correctly parsed by the Twilio voice gateway.
• Mapping: Since Hume webhooks do not pass the call_sid, maintain a server-side dictionary to map config_id to your LangGraph thread_id to ensure proper state restoration.