distill-memory

To capture durable knowledge, you can propose facts, lessons, or preferences for inclusion in the knowledge corpus. The system ensures they are sourced, classified into the appropriate memory layer, and tagged with provenance and freshness before storage.

Preventing transient state requires a validation process that proposes only durable knowledge candidates. The system avoids in-flight state and waits for operator approval before final inclusion, ensuring only verifiable facts are stored.

Operator approval works by requiring human confirmation before any proposed knowledge is added to the corpus. The system classifies candidates into memory, lessons, or user models, records their origin, and pauses for operator validation.

Yes, you can classify captured knowledge into different memory layers. The system categorizes proposed knowledge specifically into memory, lessons, or user model layers to maintain accuracy and appropriate context.

Knowledge storage needs provenance and freshness tags to ensure long-term accuracy and verifiability. Tagging each piece of knowledge with its origin and recency helps maintain a reliable, durable knowledge corpus across sessions.

To avoid duplicating facts, the knowledge capture process evaluates proposed candidates against existing entries. It prevents duplication by validating durability and sourcing before submitting the knowledge for operator approval.