Sonnet
4.5
What should we talk about today?
Identity
Your name
Instance name
Model
AI model for this profile
Use Haiku (testing mode)
Use a faster, cheaper model for testing
Extended thinking
Lets the AI think step-by-step before responding.
Thinking depth
How long the AI can think before responding
Conversation window
How much recent conversation the AI sees at once
All profiles
Retrieval sensitivity
How closely memories must match to get shown to AI. Lower is broader.
Memory retrieval limit
How many past messages get surfaced per response
Topic profiles per turn
How many people and topic profiles get loaded each turn
Concepts per turn
How many keyword-triggered background entries load per response
Recent files shown
How many recent file attachments the AI is aware of
Retrieval weights
Controls how memories are ranked when retrieved. Meaning = how closely they match what you're saying, Importance = how significant they were, Recency = how recent, Topics = whether they involve the same people or subjects.
Deep archive age
Messages older than this require @recall --deep to search
Profile settings
Features
Narrative concepts
Background knowledge that activates based on keywords
Topic profiles
Remembered profiles of people and topics
File attachments
Images, PDFs, text, and JSON files
Text-to-speech
Read responses aloud. Requires an ElevenLabs API key.
Timeline
Daily summaries of your recent conversations, last 7 days
AI Notes
Notes the AI keeps about you and ongoing topics
Context budgets
Recent
50,000 tokens
Memory
12,000 tokens
Entity
6,000 tokens
Concept
3,000 tokens
Memory tiers
Active
247 messages
Standard
1,205 messages
Deep archive
89 messages