Memory & Knowledge

AI-Parrot separates two concerns that often get confused: memory (conversation state for a given session) and knowledge (long-lived facts, documents, embeddings used by RAG).

What lives here

Memory

• ConversationMemory — abstract base for per-session histories.
• InMemoryConversation — process-local store, ideal for tests.
• RedisConversation — production-grade, multi-process safe, TTL-aware.
• EpisodicMemoryStore — FAISS-backed long-term episodic memory for agents that need to remember across sessions.
• AnswerMemory — cache of (question → answer) pairs for repeat queries.

Knowledge

• Vector stores — AbstractStore plus concrete back-ends: PgVectorStore, FaissStore, MilvusStore, ArangoStore, BigQueryStore. PgVector is the recommended default.
• parrot.knowledge — context assembly: takes a user query and returns ranked, deduplicated, citation-ready chunks.

Decision matrix

Use case	Component
Track the last N user turns	RedisConversation
Remember a user across sessions	EpisodicMemoryStore
Avoid recomputing identical queries	AnswerMemory
Semantic retrieval over documents	PgVectorStore + knowledge

Read next

• LLM Wiki — an agent-maintained knowledge repository
• Execution Memory
• Local Knowledge Base
• Parent-Child Retrieval
• DocumentDB, Storage Backends

API reference

• API Reference → Memory
• API Reference → Stores