How it works
Organization memory is a persistent key-value store shared across your org. Any user can read or write entries, and the assistant proactively uses stored context to give better answers. The assistant:- Reads memory at the start of conversations to understand your org’s context
- Writes memory when it learns useful patterns, preferences, or configurations
- Asks for confirmation before storing anything
What to store
Key addresses
Treasury wallets, contract addresses, multisigs
Preferences
Preferred chains, default time ranges, chart styles
Protocol context
Which protocols your org tracks, naming conventions, internal terminology
Patterns
Common query patterns, known data caveats, reusable SQL snippets
Example usage
Storing context
treasury_wallet→0x742d35Cc6634C0532925a3b844Bc9e7595f2bD28focus_area→DeFi on Ethereum and Arbitrum - Uniswap, Aave, Compoundour_pools_definition→Uniswap v3 pools where treasury wallet is LP
Recalling context
Using memory in analysis
Once context is stored, the assistant uses it proactively:0x742d35... without you specifying the address again.
Limits
| Limit | Value |
|---|---|
| Max entries per org | 50 |
| Max value size | 10 KB per entry |
| Scope | Organization-wide (all users share the same memory) |
What is NOT stored
- Sensitive personal data - personal wallets, credentials, private keys
- Session-specific context - the assistant already remembers everything within a conversation
- Data meant for one user - memory is org-wide, not per-user