Key Concepts
Essential terminology and concepts used in Goldilocks
Understanding these key concepts will help you get the most out of Goldilocks.
Core Concepts
RAG (Retrieval-Augmented Generation)
RAG is the technology that powers Goldilocks. When a contact asks a question:
- Retrieval - Goldilocks searches your knowledge base for relevant content
- Augmentation - The relevant content is added to the AI's context
- Generation - The AI generates a response based on your actual content
This ensures responses are grounded in your documentation, not made up.
Embeddings
Embeddings are numerical representations of text that capture meaning. Goldilocks uses embeddings to:
- Find semantically similar content
- Match contact questions to relevant answers
- Understand context beyond keyword matching
Tenants
Goldilocks is a multi-tenant platform. Each tenant (your account) has isolated:
- Knowledge base content
- Personas and settings
- Conversation history
- Analytics data
Knowledge Base Components
Documents
Documents are pieces of content your AI learns from. They can be:
- Manual text entries
- Uploaded files (PDF, DOCX, TXT)
- Crawled web pages
Documents are automatically chunked and embedded for retrieval.
FAQs
FAQs (Frequently Asked Questions) are curated question-answer pairs. They provide:
- Precise answers to common questions
- Higher priority in retrieval than general documents
- Direct control over specific responses
Document Chunks
Long documents are split into smaller chunks for better retrieval. Each chunk:
- Contains approximately 400 characters
- Has slight overlap with adjacent chunks for context
- Gets its own embedding for search
AI Configuration
Personas
Personas are AI agent configurations that define:
- Tone - How the AI communicates (friendly, professional, etc.)
- Behavior - Response length, escalation threshold
- Instructions - Custom rules and knowledge
You can have multiple personas for different use cases.
Custom Instructions
Custom instructions are additional guidance you provide to the AI:
- Business-specific rules
- Tone guidance
- Topics to avoid
- Special handling instructions
Workflows
Workflows are automated actions triggered during conversations:
- Webhooks - Send data to external systems
- Escalation - Hand off to human support
- Signals - Automatically categorize conversations
Skills
Skills are powerful integrations your AI can execute during conversations:
- Data Retrieval - Fetch real-time info like order status
- Automated Actions - Perform transactional tasks like issuing refunds
- Action Approvals - Human-in-the-loop sign-off for sensitive actions
- Logical Conditions - Granular control over when skills are triggered
Analytics Concepts
Conversation Stats
Each conversation tracks:
- Status - Active, ended, dormant
- Duration - Time from first to last message
- Outcome - Resolved, escalated, unknown
Signals
Signals categorize conversations by topic or intent:
- Help identify common issues
- Track trends over time
- Inform content improvements
Rundowns
Rundowns are AI-generated summaries of recent conversations:
- Daily, weekly, or monthly frequency
- Highlight trends and issues
- Suggest actions to improve support
Widget Concepts
Embed Code
The embed code is a JavaScript snippet that loads the Goldilocks widget on your website. It includes your API key and configuration.
Contact Context
Contact context is data passed to the widget about the current user:
- Contact ID for memory persistence
- Custom attributes for Context
- Session information
Escalation
Escalation is the process of handing a conversation to human support:
- Can be triggered by the AI, contact, or workflow
- Sends conversation transcript via email
- Includes contact form for contact details