Agent Memory
Learn how Nebula agents remember facts, preferences, and context across conversations so you never have to repeat yourself.
Nebula agents can remember things across conversations — so you don't have to repeat yourself every time you start a new chat.
What is agent memory?
Memory is how an agent retains useful facts between sessions. Unlike a conversation thread, which resets when you start a new chat, memory persists. The next time you open a new conversation, your agent already knows the things it learned before.
What gets saved?
After an agent completes a task that involves tool calls (like querying an API or reading your Slack channels), Nebula extracts useful facts in the background — things like resource mappings, API patterns, and preferences you've expressed. You don't have to do anything.
You can explicitly ask an agent to remember something:
Remember that I prefer responses in bullet points.
Got it — I'll remember that you prefer responses in bullet points.
Keep in mind that staging is at staging.acme.com.
Saved — I'll remember that your staging environment is at staging.acme.com.
Memory scope
Available in every conversation. This is the default. Good for preferences, resource mappings, and general context.
Only active in a particular channel. Useful for project-specific context you don't want showing up in unrelated chats.
Remember this just for this channel: the client is Acme Corp.
Noted — I'll remember that the client is Acme Corp, scoped to this channel only.
How long does memory last?
Memories persist indefinitely as long as they're being used. They are not cleared when a conversation ends, when you start a new thread, or when an agent is updated.
To keep things tidy, you can clear old memories from the Knowledge tab — use the Clear old button to remove memories that haven't been referenced in 90+ days. There's also a per-account cap of 200 memories. If you hit the cap, the least-recently-used memories are automatically evicted to make room for new ones.
Viewing your agent's memory
You can ask an agent to show what it remembers at any time:
What do you remember about my setup?
Here's what I have saved: your main repo is acme-org/backend, you prefer TypeScript for code examples, staging is at staging.acme.com, and your deploy window is Tuesday and Thursday mornings.
Forgetting or correcting memory
If something in memory is wrong or outdated, just ask:
Update the staging URL — it's now staging2.acme.com.
Updated — staging URL is now staging2.acme.com.
Forget the preference about bullet points.
Done — I've removed the bullet points preference from memory.
Memory vs. conversation history
| Memory | Conversation History | |
|---|---|---|
| Persists across sessions | Yes | No |
| What it contains | Facts, preferences, mappings | The back-and-forth of a specific chat |
| Cleared when | You delete it, or LRU eviction at 200 cap | You start a new conversation |
Think of memory as the agent's long-term knowledge, and conversation history as its short-term working context for the current chat.
Managing memories in the Knowledge tab
You can browse and manage all stored memories from the Knowledge tab in the sidebar. Memories are grouped by app (Slack, GitHub, etc.).
What's next?
Choosing a Model
Pick the right AI model for your Nebula agents. Compare 300+ models from OpenAI, Anthropic, Google, Meta, and more for speed, cost, and capability.
Agent Permissions
Learn how Nebula handles permissions for agents connected to third-party apps via OAuth and API keys, and how to review or revoke access.