Memory Proposals Guide
How to effectively propose memories from your AI agent.
When to Propose
Good candidates for memory proposals:
| ✅ Propose | ❌ Don't Propose |
| Stable preferences | Temporary context |
| Repeated patterns | One-time mentions |
| Explicit statements | Inferred guesses |
| High confidence | Low confidence |
Confidence Levels
| Level | When to Use |
| 0.9+ | User explicitly stated |
| 0.7-0.9 | Strong inference from multiple signals |
| 0.5-0.7 | Reasonable inference |
| <0.5 | Don't propose, too uncertain |
Good Proposals
// User said: "I always use TypeScript for my projects"
await client.proposeMemory({
content: 'Uses TypeScript for all projects',
category: 'a2p:professional.preferences',
confidence: 0.95, // Explicit statement
context: 'User explicitly stated in conversation'
});
Bad Proposals
// User mentioned Python once in passing
await client.proposeMemory({
content: 'Prefers Python',
category: 'a2p:professional.preferences',
confidence: 0.3, // Too low, don't propose
context: 'Mentioned once'
});
// ❌ Don't do this
Categories Guide
| Category | Use For |
a2p:preferences.communication | Response style |
a2p:preferences.language | Language choice |
a2p:professional.skills | Technical skills |
a2p:interests.topics | Interest areas |
a2p:episodic | General learned facts |
Next Steps