Memory

Memory gives agents the ability to remember information across conversations. Unlike knowledge (RAG), which is based on uploaded documents, memory is created dynamically during conversations and persists for future use.

How Memory Works

Conversation happens → Agent stores a memory → Future conversation → Agent recalls memory

Memory entries are embedded with vector embeddings and retrieved via semantic similarity search, similar to RAG but for agent-generated knowledge.

Memory Operations

Operation	Description
Store	Save a fact, preference, or observation for later
Recall	Search stored memories by semantic similarity

Memory Scopes

Scope	Description	Use Case
agent	Shared across all threads for this agent	User preferences, learned facts
thread	Specific to one conversation thread	Context within a session
resource	Tied to a specific external resource	Per-project or per-document notes

Configuration

Memory is configured per-agent in the Flapjack dashboard under Agent → Memory:

Setting	Description
enabled	Turn memory on/off
auto_collect	Automatically extract and store memories from conversations
auto_inject	Automatically recall relevant memories at the start of each turn
inject_count	Number of memories to inject per turn
max_memories	Maximum stored memories for this agent
agent_scope	Enable agent-wide memory
thread_scope	Enable per-thread memory
resource_scope	Enable per-resource memory

Example Flow

1. User tells the agent: "I prefer TypeScript over Python"
2. Agent stores this as a memory (automatically or explicitly)
3. In a future conversation, the agent recalls this preference
4. Agent tailors its response to use TypeScript examples

Memory vs Knowledge

	Memory	Knowledge (RAG)
Source	Created during conversations	Uploaded documents
Scope	Agent, thread, or resource	Org, agent, or thread
Content	Short facts and preferences	Document chunks
Created by	Agent (auto or explicit)	User (upload)
Best for	Personalization, learned context	Reference material, FAQs

Next Steps

• Knowledge — document-based RAG
• Agents — agent configuration
• Tools — other agent capabilities