Flowly
Log inDownload
Use cases/Knowledge & Research

Second Brain

Text any thought, link, or quote to your bot. Search semantically months later by what you meant, not what you typed.

Knowledge & Researcheasy~5m setup
Tools
memorymemory_search
Channels
telegramiosandroid

Notes apps fail because organising them is its own job. The second-brain pattern flips it: capture frictionlessly via chat, defer organisation to the search layer. Type a thought, paste a link, dictate a voice note — your bot stores it. Months later, ask in plain language and it returns the relevant memories regardless of how you phrased them originally.

What it does

  • Send anything to your bot — text, voice, links, screenshots
  • Stored in semantic memory with timestamp + source channel
  • Query in natural language; semantic match handles synonyms and paraphrasing
  • Tag-friendly: prefixing a message with #tag adds metadata for filtering
  • Voice messages auto-transcribed before storage

What you'll need

  • Just a connected Telegram (or any) channel
  • Memory tool — built-in, on by default

Setup

1. Set the capture intent

Send your bot once:

Send to Flowly
Default rule: when I send you a fragment that's clearly a note (no question, no command, no greeting), store it in memory with whatever tags make sense. Don't reply with anything more than ✓ unless I'm asking a question.

That's it. From now on, casual messages just get logged.

2. Capture habits

Some patterns that work:

  • Quotes: paste a passage, prefix "quote: " — "quote: 'The map is not the territory.' — Korzybski"
  • URLs: paste with one-line context — "todo read this when free: https://...". The bot extracts page title automatically.
  • Voice notes: hold mic button, talk freely. Transcribed + stored.
  • Tags: any #word becomes searchable — #book #startup-idea #meeting-prep.

Type what you remember, however you remember it:

"What did I save about agent architectures?" "That quote about maps and territory" "Things I tagged #startup-idea last month"

The agent searches memory semantically (not literal substring), pulls top 5 matches, summarises each.

Tips

  • Trust the search. The temptation is to hand-organise into notebooks/folders. Don't. Semantic search is good enough that organisation overhead exceeds retrieval friction.
  • Capture in the moment. The friction window is ~10 seconds — longer than that and you forget the thought. Voice notes during a walk catch ideas you'd lose at a keyboard.
  • Periodic reviews. Once a month ask: "Show me what I saved this month, grouped by topic." Patterns emerge that you wouldn't notice from inside individual notes.
  • Prune aggressively. When the agent surfaces something 3+ months old that's clearly no longer relevant, ask it to delete. Memory bloat is real and it slows search.
  • Cross-channel works. Capture from iOS by voice while walking. Search from Desktop while writing. Same memory, no sync overhead.