Episodic Memory

You have access to a searchable index of all past pi conversations across all projects.

When to Search

• User says "like we did before", "remember when", "how did we solve X", "what did we try"
• User references a past session, project, or decision
• You encounter an error or pattern that might have been addressed previously
• Starting work on a feature similar to something done before
• User asks about project history or past decisions

How to Search

Use the episodic_memory_search tool:

// Semantic search (finds by meaning, not just keywords)
{ "query": "authentication middleware implementation" }

// Multi-concept AND search (results must match ALL concepts)
{ "query": ["React Router", "authentication", "JWT"] }

// Filter by date
{ "query": "database migration", "after": "2025-01-01" }

// Text search for exact phrases
{ "query": "useAuth hook", "mode": "text" }

Tips for Good Queries

• Use descriptive phrases, not single keywords
• Try the problem description, not just the solution
• If first search doesn't find it, try rephrasing or broadening
• Use multi-concept search when looking for a specific intersection of topics

Viewing Full Conversations

If a search result snippet isn't enough context, use episodic_memory_show with the session file path from the search results to see the full conversation.

Important

• Search results are snippets (chunks of ~4 turns). They show the relevant part, not the whole conversation.
• Results include a relevance score — higher is better.
• The project name in results shows which directory the session was in.
• Don't search for every query — only when past context would genuinely help.