Agent Memory

Nibchat includes a built-in persistent memory system. The agent can save, list, and delete free-form notes about the user — and those notes are automatically injected into every future conversation.

How it works

Memory is enabled automatically — no configuration required.

1. Auto-injection — At the start of each conversation, all saved memories for the session are loaded and injected into the system prompt.
2. Proactive saving — The agent is instructed to proactively call save_memory whenever it learns something useful about the user (preferences, context, facts).
3. Silent tool calls — Memory tool calls (save_memory, delete_memory, list_memories) do not appear as visible tool-call cards in the chat UI. They happen silently in the background.

Memory tools

The agent has access to three built-in memory tools:

Tool	Description
save_memory	Save a free-form note to memory.
delete_memory	Delete a memory by its ID.
list_memories	List all saved memories for this session.

Scope

Memories are per-session. Each anonymous user session has its own isolated memory store. Memories persist indefinitely until explicitly deleted.

User interaction

Users can interact with memory through natural language:

• "What do you remember about me?" — triggers list_memories
• "Forget that I prefer dark mode" — triggers delete_memory
• "Remember that I'm a developer" — triggers save_memory

Example

After a conversation where the user mentions they prefer concise answers and work in Python:

Memory store:
- [id:1] User prefers concise, direct answers without unnecessary filler.
- [id:2] User primarily works in Python.

In the next conversation, the system prompt will include:

## What you remember about this user
- [id:1] User prefers concise, direct answers without unnecessary filler.
- [id:2] User primarily works in Python.

The agent uses this context automatically without the user needing to repeat themselves.