v3.2.0

Release Notes

New: Chat Turn Sensors (per user/device)

You now get per-user and per-device sensors that count conversation turns. These sensors are grouped under the xAI Notifications device (same group as sensor.xai_notifications_available_models). They let users decide how to handle long chats via automations without any forced resets.

Sensor names

• <User Name> - <Conversation Subentry Title> (<mode>) (e.g., Alberto - xAI Conversation (pipeline))
• <Device Name> - <Conversation Subentry Title> (<mode>) (e.g., Kitchen - xAI Conversation (tools))

What counts as a “turn”

A turn is stored when a conversation response ID is saved in memory. The sensor value aggregates all stored turns for that user/device (across modes and subentries).

---

Example Automation: Reset Long Chats (User/Device Controlled)

This example uses the Clear Conversation Memory service to reset chat history when a user’s turn count exceeds a threshold. It’s optional and fully controlled by the user.

alias: Reset chat memory when turns exceed 60
mode: single
trigger:
  - platform: numeric_state
    entity_id: sensor.alberto_xai_conversation_pipeline
    above: 60
action:
  - service: xai_conversation.clear_memory
    data:
      user_id:
        - person.alberto
      include_server: true
      only_inactive: false

Optional: include inactive sessions only

Use this to clear only conversations that are already inactive (e.g., older than the configured TTL).

alias: Clear inactive chat memory for a device
mode: single
trigger:
  - platform: numeric_state
    entity_id: sensor.kitchen_xai_conversation_tools
    above: 100
action:
  - service: xai_conversation.clear_memory
    data:
      satellite_id:
        - assist_satellite.kitchen
      include_server: true
      only_inactive: true

Optional: clear everything

Use with caution.

alias: Clear all chat memory (manual)
mode: single
trigger: []
action:
  - service: xai_conversation.clear_memory
    data:
      delete_all: true
      include_server: true

---

Also includes internal optimizations across intent pipeline, fallback handling, memory management, prompt construction, and tool orchestration to improve overall stability, safety, and responsiveness.

---

Full Changelog: 3.1.0...3.2.0