knowledge-agent

You build a queryable knowledge base from observation history by filtering past records by project, types, concepts, file prefixes, semantic query, or date ranges to compile a focused, reusable corpus for analysis.

Yes, you can filter past observations for retrospective analysis using semantic query and date range limits, alongside project, types, concepts, and file prefixes, to isolate specific work patterns and prior decisions into a compiled corpus.

The best way to query past decisions across a time window is to compile filtered observations into a corpus, prime it into an AI session, and use conversational querying to retrieve context-aware answers about historical outcomes.

Conversational querying works by priming a compiled expertise corpus into an AI session, enabling you to ask follow-up questions with consistent context and optionally refreshing the session when context drifts.

Yes, you need existing claude-mem observation data because the corpus is built by filtering and compiling those specific past observations into a queryable knowledge base for retrieval augmented recall.