Skip to main content

Conversation Checkpoints

Ever had that "oh no" moment when your AI conversation goes sideways? Maybe you asked a question that spawned 47 follow-ups, the AI got confused, or you just want to travel back in time to before things got... complicated.

Welcome to Conversation Checkpoints. It's like Ctrl+Z, but for AI conversations.

The Problem We're Solving​

Picture this: You've built up a beautiful analysis over 20 messages. Your AI understands your data, knows what you're looking for, and then... you ask one wrong question. Suddenly the context is polluted, the AI is confused, and starting over means losing everything.

Traditional solution: Cry. Start a new conversation. Re-explain everything.

QRY solution: Just rewind. Like nothing happened.

How It Works​

Automatic Checkpoints​

QRY automatically creates checkpoints before each message you send. Think of it as auto-save, but smarter:

  • Zero effort required: Checkpoints happen automatically in the background
  • Smart cooldown: We don't spam your conversation with checkpoints (5-second minimum between saves)
  • Lightweight: Checkpoints are tiny - they don't bloat your data

Manual Checkpoints​

For those "this is important, let me save right now" moments:

  1. Click the checkpoint icon (⟲) in your conversation header
  2. Click "Create Checkpoint"
  3. Give it a memorable name (like "Before I ask about Q4 numbers")

Rewinding​

When things go wrong:

  1. Click the checkpoint icon
  2. Browse your checkpoints (newest first)
  3. See exactly what will be removed
  4. Click to rewind
  5. Watch the magic happen

The removed message even gets restored to your input field - so you can rephrase and try again.

What Actually Happens​

When you rewind:

  1. Messages after the checkpoint: Softly removed (we keep them in case you change your mind... okay, mainly for database integrity, but still)
  2. AI context: Completely cleared. The AI genuinely forgets everything after that point
  3. Your input: The question that caused the mess? It's back in your text box, ready for a second attempt

Pro Tips​

Best Moments for Manual Checkpoints
  • Before asking a potentially complex question
  • When your analysis is "just right" and you want to preserve it
  • Before letting an intern use your conversation (kidding... mostly)
Checkpoint Limits

We keep up to 50 checkpoints per conversation and automatically clean up older ones after 30 days. Manual checkpoints with labels are preserved longer than automatic ones.

The Technical Magic (For the Curious)​

Behind the scenes, rewinding:

  • Soft-deletes messages (keeps data integrity happy)
  • Clears Redis session caches (so the AI truly forgets)
  • Invalidates LLM service memory (belt AND suspenders approach)
  • Deletes future checkpoints (because time travel creates paradoxes)

It's like we hit the AI with a targeted amnesia ray. Very sci-fi. Very effective.

FAQ​

Q: Can I recover messages after rewinding? A: The messages are soft-deleted, so technically yes - but you'll need to ask your admin nicely.

Q: Will rewinding affect my other conversations? A: Nope. Each conversation is independent. Time travel is localized.

Q: How far back can I go? A: As far as your earliest checkpoint. We keep up to 50 per conversation.

Q: Does this work with all AI models? A: Yes! Claude, Gemini, GPT - they all get the same "forget this ever happened" treatment.

Because sometimes the best way forward is to go back. ⟲

QRYA product of IXEN.