Memory
Your assistant remembers preferences and facts across workflow runs using the memory system.
How it works
When you tell your assistant something like “always include weather in my briefings,” it stores that as a memory. The next time the relevant workflow runs, it has access to that memory.
Memory types
Global memories
Apply to all workflows. Use these for preferences that affect everything your assistant does.
Examples:
- “My assistant’s name is Alex”
- “I prefer bullet points over paragraphs”
- “Always use formal greetings in emails”
- “My work hours are 9 AM to 6 PM Pacific”
Per-workflow memories
Apply to a specific workflow only. Use these for preferences that only matter in a particular context.
Examples:
- “Include SF weather in morning briefings” → Morning Briefing only
- “Always CC my assistant on scheduling replies” → Schedule Assist only
- “Don’t draft replies to recruiters” → Auto-draft only
Adding memories
Ask your assistant from any channel:
- “Remember to always include the weather in my briefings”
- “From now on, keep Friday afternoons free for focus time”
- “When drafting emails, always sign off with just my first name”
Your assistant determines whether the memory should be global or workflow-specific based on context.
Viewing and managing memories
In the web app:
- Go to Settings → Profile
- View all stored memories
- Delete any you no longer want
Or ask your assistant:
- “What memories do you have stored?”
- “Forget the preference about Friday afternoons”
How workflows use memories
When a workflow runs, it receives:
- All global memories
- Any memories specific to that workflow
The workflow’s prompt has access to these memories and incorporates them into its behavior.
Memory tools
Workflows can read and write memories using these tools:
| Tool | Description |
|---|---|
get_memories | Retrieve stored memories |
add_memory | Store a new memory |
delete_memory | Remove a stored memory |
// Add a workflow-specific memory
add_memory({
content: "Always include weather for San Francisco",
agent_slug: "town-morning-briefing"
})
// Add a global memory (omit agent_slug)
add_memory({
content: "I prefer concise responses"
})