Assistants Explained

Assistants are AI-powered agents in Gen AI Builder, part of EDB Postgres® AI (EDB PG AI). They unify language understanding, knowledge retrieval, behavior control, and task execution to create powerful, governed AI applications.

Assistants enable you to build:

• Sovereign AI experiences — using your data, on your infrastructure
• AI copilots and agents with controlled behavior and trusted knowledge
• Conversational apps integrated with Postgres and Hybrid Manager

Assistants are the primary deployment target for production AI applications in EDB PG AI.

Before you start

You’ll get the most out of this section if you have:

• Basic understanding of Large Language Models (LLMs)
• Familiarity with Knowledge Bases Explained, Retrievers Explained, Rulesets Explained, Tools Explained, and Memory concepts in AI Factory
• Helpful: experience with Hybrid Manager and Model Serving

Suggested starting points:

• AI Factory Concepts
• AI Factory 101 Path

What is an Assistant?

An Assistant is an AI agent that interacts with users and systems, performing tasks based on:

• Natural language input and generation
• Retrieved, trusted knowledge
• Predefined behavioral rules
• Conversation memory
• External tools and APIs

Assistants combine:

Without Assistants, your AI Factory content — Knowledge Bases, Rulesets, Tools — remains static. Assistants activate and unify these components into working applications.

Why use Assistants?

• Build AI-driven user experiences — chatbots, copilots, interactive agents
• Combine retrieved knowledge with controlled behavior
• Enable multi-step reasoning and task execution
• Enforce compliance, brand tone, and governance
• Provide conversation memory and personalized context
• Run Sovereign AI — your data, governed behavior, full observability

How Assistants work

At runtime:

1. User submits input (text, API call, or other supported modality)
2. Assistant pipeline executes:

User Input → Rulesets → Retriever → Knowledge Bases → Retrieved Content
→ Tools → Action Execution
→ Memory → Context Maintenance
→ LLM → Response → User Output

• Rulesets define behavioral policies and tone
• Retriever executes semantic search on assigned Knowledge Bases
• Tools trigger external actions if required
• Memory injects conversation history and state
• LLM synthesizes and generates final response

Core components of an Assistant

LLM

• The Large Language Model used for understanding and generation
• Examples:
• GPT-4
• Claude 3 Sonnet
• Gemini Pro
• Custom models deployed via Model Serving

Knowledge Bases + Retriever

• Retrieval-Augmented Generation (RAG) pipeline:
• Knowledge Bases: trusted, indexed content
• Retrievers: semantic search behavior and tuning

Rulesets

• Define how the Assistant behaves:
• Tone and style
• Policy enforcement
• Compliance requirements
• Branding guidelines

Tools

• Extend Assistant capabilities:
• API integrations
• Calculators, lookups
• Internal system actions
• External data fetchers

Memory

• Maintains conversation context:
• Number of conversation turns remembered
• Summarization or truncation policies
• Supports multi-turn reasoning and personalization

When to use Assistants

Use Assistants when you want to expose AI Factory capabilities through:

• Conversational interfaces (chatbots, support bots, knowledge advisors)
• Internal productivity tools (policy checkers, knowledge agents)
• Copilot-style assistants (embedded in business applications)
• Custom API endpoints driven by AI logic

Assistants are the primary vehicle for production AI applications in Gen AI Builder.

Patterns of use

Single Assistant per use case

• One Assistant → one primary interaction pattern
• Example: Postgres Migration Copilot AI

Persona-based Assistants

• Different Assistants for different personas or business roles
• Examples:
• Financial Advisor Copilot
• Internal Compliance Agent
• Marketing Content Generator

Shared Knowledge + Ruleset Assistants

• Multiple Assistants sharing Knowledge Bases and Rulesets
• Enforce consistent behavior and compliance across applications

Memory-driven Assistants

• Assistants configured with advanced Memory settings
• Ideal for:
• Long-term user relationships (e.g., CRM agents)
• Complex multi-turn workflows (e.g., troubleshooting bots)

Best practices

• Always combine Knowledge Bases + Rulesets → for grounded, controlled behavior
• Use Tools sparingly — only where needed
• Start simple → iterate on:
• System prompt
• Rulesets
• Retriever tuning
• Memory configuration
• Test Assistants extensively before production
• Monitor performance and refine continuously
• Govern and version Rulesets and Knowledge Bases — critical for Sovereign AI

Governance and Sovereign AI

Assistants in EDB PG AI enable Sovereign AI principles:

• Data stays in your databases and infrastructure
• Model serving is powered by your infrastructure via Model Serving
• Knowledge Bases, Rulesets, and Tools are fully auditable and versioned
• Full observability is provided through Hybrid Manager AI Factory dashboards

Assistants are designed for production-grade, compliant, governed AI.

Related topics

• Create an Assistant
• Working with Assistants
• Rulesets Explained
• Knowledge Bases Explained
• Retrievers Explained
• Structures Explained
• AI Factory Concepts
• Hybrid Manager AI Factory

Next steps

• Explore other AI Factory Explained pages
• Start building your first Assistant:
• AI Factory 101 Path
• How-to: Create an Assistant