Memory

Give your agents persistent memory so they can remember conversations, learn from past interactions, and provide personalized experiences.

What is Memory?

By default, AI models are stateless — they forget everything after each response. Fluxgate Memory gives agents the ability to remember information across multiple runs and sessions.

Memory Types

Type	Duration	Use Case
session	Within a conversation	Chat history, context within current interaction
long_term	Permanent	User preferences, facts, learned information
episodic	Event-based	Past interactions, "I remember when you asked about..."
semantic	Permanent (vector)	Meaning-based search using embeddings (pgvector)

Store a Memory

Store Memorybash

curl -X POST http://localhost:8000/api/v1/memory \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "agent_id": "AGENT_ID",
    "memory_type": "long_term",
    "content": "User prefers email communication over phone calls",
    "importance": 0.8,
    "metadata": { "user_id": "user_123", "source": "conversation" }
  }'

Search Memory

Use semantic search to find relevant memories based on meaning, not just keywords:

Search Memorybash

curl -X POST http://localhost:8000/api/v1/memory/search \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "query": "communication preferences",
    "agent_id": "AGENT_ID",
    "memory_type": "long_term",
    "limit": 5
  }'

List Memory Entries

List Memorybash

# List all memories for an agent
curl "http://localhost:8000/api/v1/memory?agent_id=AGENT_ID" \
  -H "Authorization: Bearer YOUR_API_KEY"

# Filter by type
curl "http://localhost:8000/api/v1/memory?agent_id=AGENT_ID&memory_type=session" \
  -H "Authorization: Bearer YOUR_API_KEY"

Vector Search with pgvector

Semantic memory uses pgvector under the hood. When you store a memory with type "semantic", Fluxgate automatically generates an embedding vector and stores it for similarity search. This means you can find related memories even if they use different words.

Using the Dashboard

The Memory tab in the dashboard lets you:

Browse all memory entries across all agents
Filter by memory type (session, long_term, episodic, semantic)
Search memories by content or metadata
View importance scores and timestamps
Delete individual memories or bulk clear

Best Practices

Use session memory for conversation context (auto-managed)
Use long_term for facts that should persist forever
Set importance scores (0-1) to help prioritize relevant memories
Clean up old session memories periodically to save storage
Use semantic memory when you need fuzzy/meaning-based retrieval

Tools

Scheduling