First run: If slow on first use, run

bunx memory-search --warmup

to pre-download the embedding model (~300MB).

• path: File where the match was found
• lines: Line range (startLine-endLine)
• score: Relevance score (0-1)
• snippet: The matching text

1. Present the relevant results to the user
2. If they want more detail, use

   Read

   to get the full file content
3. If no results found, offer to search with different terms

• User preferences ("I prefer TypeScript")
• Project decisions ("We chose PostgreSQL for X reason")
• Coding patterns ("Always use async/await")
• Architecture decisions
• Important URLs, contacts, credentials references

• What was worked on today
• Bugs found and fixed
• Ideas to explore later
• Meeting notes
• Temporary context

1. Read existing file first (if it exists) to avoid overwriting
2. Append new content with a timestamp or section header
3. Use clear, searchable language (will be vector-searched later)
4. Run sync after saving to update the search index

bunx memory-search --sync

• Prefers Bun over Node for TypeScript projects
• Uses pnpm as package manager
• Likes minimal dependencies

• Single-user application
• No need for concurrent writes
• Simpler deployment

undefined

• Implemented JWT auth flow
• Added refresh token rotation
• TODO: Add rate limiting

• Fixed race condition in user creation
• Root cause: missing transaction wrapper

---

• Use descriptive queries, not single keywords
• The search is semantic (understands meaning, not just exact words)
• If unsure which file to use, ask the user
• Always sync after saving new content