Procedural Memory Provider

Overview

The ProceduralMemoryProvider manages user-specific behavioral data. Unlike Episodic memory (which is about what happened), Procedural memory is about who the user is and how the bot should interact with them.

Functionality

The provider fetches UserInfo from the SQLiteUserManager. This data typically includes: - User Bio: A short description of the user. - Interests: Topics the user cares about. - Custom Instructions: Specific rules the user has set for the bot's behavior towards them. - Interaction History: High-level summaries of past interactions.

Technical Implementation

TTL Cache

To avoid hitting the SQLite database for every single message, the provider uses a per-user cache: - Logic: Each user_id is cached with its own expiration timer. - Invalidation: The invalidate(user_id) method is called by the /memory save command to ensure that updates are reflected immediately.

Batch Fetching

The provider supports get_multiple_users(), allowing it to fetch context for everyone involved in a multi-user conversation in a single pass.

Schema: UserInfo

The data is structured using Pydantic models (defined in llm.memory.schema): - user_id: Discord ID. - nickname: User's display name. - bio: Extracted or user-provided biography. - instructions: Custom behavioral overrides.

Integration

The result is injected into the prompt as a dedicated section:

--- User Profiles ---
User: Alice
Bio: Loves coding and cats.
Instructions: Always be polite.
---

---

Procedural memory is the foundation of the bot's personalized interaction model, allowing it to adapt its tone and knowledge to each individual user.