Memory Basics

Goal: Enable agents to remember information across conversations and searches.

Prerequisites

Complete orchestration-basics.md to understand multi-agent systems.

What is Memory?

Memory allows agents to:

• Remember previous conversations
• Store and search knowledge documents
• Perform RAG (Retrieval-Augmented Generation)
• Share context between agents

1) Start with in-memory provider

agentcli create memory-demo --memory memory --agents 2
Set-Location memory-demo

This creates a project with memory enabled. Check the generated agentflow.toml:

[agent_memory]
provider = "memory"          # In-memory storage (non-persistent)
auto_embed = true           # Automatically create embeddings
max_results = 10           # Maximum search results
enable_knowledge_base = true

Validate and run:

agentcli validate
go run . -m "Remember that I like pizza"

Expected: Agent stores the information and can reference it in responses.

2) Test memory persistence

Run another query:

go run . -m "What food do I like?"

Expected: Agent recalls your pizza preference from the previous interaction.

3) Enable RAG (Retrieval-Augmented Generation)

Edit agentflow.toml to add RAG:

[agent_memory]
provider = "memory"
enable_rag = true
chunk_size = 1000
chunk_overlap = 100
rag_max_context_tokens = 4000

This allows agents to break down large documents and find relevant sections.

4) Going persistent (PostgreSQL + pgvector)

For production use, create a persistent memory system:

Set-Location ..
agentcli create persistent-demo --memory pgvector --agents 2
Set-Location persistent-demo

Check the generated configuration:

[agent_memory]
provider = "pgvector"
connection = "postgres://user:password@localhost:15432/agentflow?sslmode=disable"
dimensions = 1536
enable_rag = true

⚠️ Note: You need a running PostgreSQL database with pgvector extension. See the Memory Provider Setup Guide for database setup instructions.

5) Alternative: Weaviate vector database

agentcli create weaviate-demo --memory weaviate --agents 2

Configuration:

[agent_memory]
provider = "weaviate"
connection = "http://localhost:8080"

Memory Provider Comparison

Provider	Persistence	Setup	Best For
memory	No	None	Development, testing
pgvector	Yes	PostgreSQL + extension	Production, SQL ecosystem
weaviate	Yes	Docker container	Production, advanced vector features

Next Steps

• Add external tools: tools-basics.md
• Learn deployment: deploy-basics.md

Verification checklist

• [ ] In-memory project created and validated
• [ ] Memory persistence demonstrated across runs
• [ ] RAG configuration understood
• [ ] Persistent provider option explored