For Operators: Memory
It learns. It remembers. It gets better.
Your AI Persona absorbs everything you give it: documents, updates, corrections, and conversations. It learns from every interaction, remembers every client, and improves over time. Memory isn't a feature. It's how an AI Persona grows.
Aria's Memory
Sales Assistant · Apex Logistics
Knowledge you added
Product Catalog 2024.pdf
4.2 MB
Pricing Sheet Q1.xlsx
890 KB
FAQ & Objection Handling.docx
1.1 MB
Service Overview Deck.pdf
6.8 MB
Learned from conversations
847
Interactions
31
Contacts
4.2k
Facts stored
Last updated 2 hours ago
01. Architecture
Four layers of memory
Memory comes from two sources: what you deliberately feed in, and what the AI Persona learns on its own. Each layer serves a distinct purpose, and together they give your Persona the ability to know, recall, relate, and discover.
Short-Term Memory
The active context held during a single conversation. It tracks everything said so far: the topic, what was asked, what was answered, and any shifts in direction. When the session ends, short-term memory is compacted, not discarded. The substance is preserved and feeds into the Persona's long-term systems.
Live session, right now
Compacted when the conversation ends. Substance preserved.
Memory Maps
Structured schemas that the AI Persona generates on the fly. When a document is uploaded or data is added, the AI Persona determines what fields should exist and builds a structured map from the content. Unstructured and structured data alike are organized into these maps, making them usable for retrieval, reasoning, and external integration.
Map generated from upload
Treated History
The relational layer. The AI Persona catalogues conversations by user and topic, building associative memory over time. When something new comes up, it connects it to loosely related past conversations without being prompted.
Associative recall
Just picked up a new car
That's great. Does that mean you moved on from the one you told me about after the crash last year?
Linked to conversation from 12 months ago
Vector Search
Broad similarity retrieval that extends beyond structured maps and conversation history. The AI Persona can search across its full knowledge base using vector similarity, either through a third-party provider or through an LLM's built-in knowledge capacity. This layer handles queries where the answer isn't neatly categorized but exists somewhere in the AI Persona's accumulated knowledge.
Similarity search
Discovery Call · Mar 12
"…would want to revisit pricing around mid-year…"
Pricing Sheet Q1.xlsx
"…renewal requires 30-day notice…"
Follow-up call · Jul 3
"…budget confirmed for Q3 renewal…"
02. Feed it knowledge
Teach it what you know.
There's no single way to build a Persona's knowledge. Upload files and it reads them. Send it updates and it absorbs them. Train it through conversation and it adapts. Knowledge isn't static. It's something you add to, correct, and refine over time.
Documents and files
PDFs, Word files, plain text: product guides, service descriptions, policies, onboarding materials, legal terms
Spreadsheets and structured data
Excel or CSV files: pricing tables, inventory lists, client records, anything with rows and columns
Presentations and images
Slide decks, pitch materials, product images, diagrams: visual context the Persona can reference and describe
Information updates
Push new information to the Persona at any time. Price changes, policy updates, new product launches, seasonal adjustments. The Persona incorporates them immediately without needing a full document upload.
Training and correction
Teach the Persona through direct interaction. Correct a response it got wrong, clarify how it should handle a scenario, tell it what to prioritize. Each interaction refines its understanding.
Live data via API
Connect structured data sources directly (inventory systems, CRMs, databases) so memory stays current automatically
Teaching Aria
Behavior refinement · Apex Logistics
Aria's response
Your correction
"Don't mention brochures. Instead, ask about the client's shipment volume and timeline first."
Personality tuning
03. Teach
Train it. Refine it. Make it yours.
Every AI Persona starts with a Core Skill, the ability to communicate naturally. What happens next is yours to shape.
Correct a response that missed the mark. Adjust the tone. Tell it what to avoid and what to lead with. Over time, the Persona absorbs your feedback and makes it permanent, like raising and educating a new hire until they represent your brand exactly as you'd want.
Correct responses directly
Flag a reply you didn't like and tell the Persona what to say instead. It learns the correction and applies it going forward.
Tune personality traits
Adjust tone, assertiveness, detail level, and warmth. Dial each trait until the Persona sounds like your brand.
Reinforce what works
Mark responses you want more of. The Persona recognises what's working and weights those patterns higher in future conversations.
04. For operators
What memory means for your business
Feed it your business knowledge
Upload your product catalog, pricing sheets, service guides, internal FAQs, legal documents, or any file your AI Persona needs to know. Push updates when things change. Train it through conversation when a document isn't the right format. It reads, absorbs, and internalizes everything so when a client asks a question, it draws on your real knowledge, not guesswork.
Contact history without a CRM
Every time a client or prospect speaks with your AI Persona, their details, questions, and shared context are stored automatically. The next time they return, days or months later, your AI Persona picks up exactly where they left off. No manual data entry. No follow-up notes to write.
Lead qualification that happens in conversation
As your AI Persona talks to prospects, it tracks signals: intent language, budget mentions, company size, urgency cues. Over multiple sessions it builds a qualification profile automatically. When a lead is ready, your AI Persona knows and can surface them without you having to review transcripts.
Lineage memory across Persona generations
When you create a child Persona from a parent, it inherits a blend of that parent's long-term memory. Proven knowledge, learned client preferences, and successful response patterns carry forward. Each new generation of AI Personas starts ahead of where the last one began.
Memory snapshot
What Aria knows after 30 days
Documents you uploaded
4 files absorbed · 12 MB total · All indexed
Client profiles built
Marcus Chen
47 facts
Priya Nair
31 facts
David Osei
19 facts