Best Practices

Use appropriate categories

Choose the right category for each memory. This improves search relevance and helps with memory organization. See Memory Categories.

Include rich metadata

Add tags, importance levels, and custom metadata to memories. This enables powerful filtering during search.

store_memory({
  content: "User prefers dark mode",
  category: "preferences",
  tags: ["ui", "settings"],
  importance: 0.8,
  metadata: { source: "settings_page" }
})

Keep memories atomic

Store one concept per memory. Don't bundle unrelated information together. This improves search accuracy and makes updates easier.

Use semantic search for recall

Semantic search finds conceptually similar memories even with different wording. Use it when you need to recall information based on meaning.

Filter by category when appropriate

When you know what type of memory you're looking for, filter by category to narrow results and improve relevance.

Tune similarity thresholds

Adjust the similarity threshold based on your needs. Lower values (0.5) return more results, higher values (0.8) return only close matches.

Set appropriate expiry times

Use expires_at for temporary memories. Context that's only relevant for a session should expire after the session ends.

Archive instead of delete

When memories become outdated, archive them instead of deleting. Archived memories can be useful for temporal queries and debugging.

Review contradictions regularly

Use check_contradictions periodically to identify conflicting memories that need resolution.

Batch operations when possible

If storing multiple related memories, consider batching them to reduce API calls and improve throughput.

Use projects for organization

Group related memories into projects. This makes management easier and can improve search performance within a scope.