Overview
Memory gives the agent persistent context about you that carries across every conversation. When you share a preference, make a decision, or tell the agent something important, it can save that information and recall it later — even in a completely different chat thread.
Memory is shared across all your conversations. A fact saved in one thread is available in every future thread.
How memory works
The agent has two memory tools:
save_memory — Stores a fact, preference, decision, or entity that should be remembered across conversations.memory_search — Searches stored memories by keyword or topic to recall relevant context.
When you start a new conversation, the agent automatically receives your most recent memories so it has context about you from the start.
Saving memories
There are two ways memories get saved:
Explicitly ask the agent
Tell the agent to remember something:
Remember that I prefer Python over JavaScript
Note that my timezone is America/Los_Angeles
Keep in mind that we decided to use PostgreSQL for the analytics project
save_memory and confirm what was saved.
The agent saves proactively
The agent is designed to recognize memory-worthy information and save it on its own. When you share durable context like:
- Preferences — “I always use Tailwind for styling”
- Decisions — “Let’s go with Next.js for the frontend”
- Personal context — “I work at Acme Corp” or “My name is Alex”
- Project facts — “Our API is at api.example.com”
The agent will save these as memories without being explicitly asked.
Recalling memories
The agent automatically recalls memories in two ways:
- At the start of every conversation — your most recent memories are injected into context, so the agent already knows your preferences.
- On demand via search — when you ask about something from a previous conversation, the agent searches its memory database.
Try asking in a new thread:
What programming language do I prefer?
What project am I working on?
Managing memories
You can view, search, filter, and delete memories from the Memory page in the sidebar.
Searching memories
Use the search bar to find specific memories by keyword. Results update as you type.
Filtering by category
Click the category pills to filter memories by type: Preferences, Facts, Decisions, Entities, or Other.
Deleting memories
Hover over any memory card and click the trash icon to delete it. The agent will no longer recall deleted memories.
Deleted memories are permanently removed. The agent will not be able to recall them in future conversations.
Memory categories
Each memory is tagged with a category:
| Category | What it stores | Examples |
|---|
| Preference | Likes, dislikes, tool/language choices | ”User prefers dark mode”, “User likes TypeScript” |
| Fact | Objective information about you or your work | ”User’s timezone is PST”, “Uses Redis for caching” |
| Decision | Choices made that affect future work | ”Decided to use PostgreSQL for analytics” |
| Entity | Names, contacts, companies | ”User works at Acme Corp”, “User’s name is Alex” |
| Other | Anything that doesn’t fit the above | General notes and reminders |
Memory and thread notes
The agent has two tiers of memory:
| Thread Notes | Cross-Thread Memory |
|---|
| Scope | Current conversation only | All conversations |
| Tool | save_thread_notes | save_memory / memory_search |
| Survives | Context compaction within the thread | Indefinitely across all threads |
| Best for | Task-specific context (file paths, current approach) | Durable facts about you (preferences, decisions, identity) |
You don’t need to manage thread notes — the agent handles them automatically. Focus on cross-thread memory for the things you want remembered permanently.
