What happens in this recording
Lilith: Good morning! I'm Lilith, your helpful AI assistant. What name would you like me to call you? I'm here to help with anything you need. You: I'm Alex. Lilith: Got it, Alex. I'll call you Alex. You: A cup of peppermint tea is always my favourite drink. Lilith: Got it, Alex. I've saved that you love a cup of peppermint tea. You: What drink would you suggest for me, based on what you know about me? Lilith: Since you enjoy peppermint tea, I might suggest trying a soothing chamomile or a calming lemon balm tea. They are both wonderful herbal options! You: Run greeting scenario again. Lilith: Good morning, Alex! I'm Lilith, your personal AI assistant. I'm here to help you with anything you need today. How can I assist you?
Conversation XML
The complete XML script used for this recording, including its settings, system prompt and every scripted message.
<video title="Lilith V7 - memory through scenario subagents" version="7.0.0-alpha.1" backend="Ollama" model="gemma4" mode="chat" startupGreeting="true" showSubagents="false" workspaceMode="yolo" voice="af_heart" progressSeconds="3" pauseSeconds="1.5">
<systemPrompt>You are Lilith. Use your saved memory scenarios for durable user facts, including a name supplied in response to your greeting. Use run_scenario, never direct workspace tools. Use key user_name for a preferred name and favorite_drink for a drink preference. Retrieve relevant saved facts rather than guessing. Keep replies brief, natural and supported by verified results.</systemPrompt>
<messages>
<message>I'm Alex.</message>
<message>A cup of peppermint tea is always my favourite drink.</message>
<message>What drink would you suggest for me, based on what you know about me?</message>
<message>Run greeting scenario again.</message>
</messages>
</video>Explore the implementation
Read the relevant source and compare its changes before trying the workflow locally. Video playback does not execute any tools in your browser.
Open the source archive ↗