Long-term Memory

Long-term memory allows an AI agent to retain and recall information from previous conversations. Unlike session-based chat that forgets everything when the conversation ends, a node with long-term memory can remember user preferences, past decisions, and contextual details across unlimited sessions.

Interlocute automatically embeds each interaction into vector storage and indexes it for semantic similarity search. When a new message arrives, the platform retrieves the most relevant past interactions and injects them into the LLM context window. This happens transparently — no code required beyond enabling the feature.

No. Memory storage is fully managed by Interlocute. There are no databases to provision, no schemas to maintain, and no backup procedures to implement. Memory persists as long as the node exists.

Yes. You can configure TTL (time-to-live) settings to control how long individual memories remain active. Expired memories move to cold storage and are no longer surfaced during semantic retrieval, but can be restored if needed.

Yes. Each node has its own isolated memory partition. Memories stored by one node are never accessible to another node, ensuring data isolation for multi-tenant deployments and distinct use cases.

Memory, RAG, tool use, and scheduling are all composable. A node can use long-term memory alongside a RAG knowledge base — the memory provides user-specific context while RAG provides document-level knowledge. Both are injected into the LLM prompt automatically.

Memory stores and retrieves context from past conversations — things the user has said or the node has learned. RAG retrieves context from documents you upload. Memory is interaction-based and grows over time; RAG is document-based and is updated when you add or change files.

Memory operations are metered as part of your node's computation usage. Each write (embedding a new interaction) and each read (semantic retrieval) is logged and included in your usage ledger. There are no separate storage fees.