Every message you send includes your entire chat history. By message 100, your chatbot is slow, expensive, and still forgetting things that matter. ChatSorter fixes that — keeping only what's actually relevant, automatically.
# After every user message requests.post("api.chatsorter.com/process", json={ "chat_id": user_id, "message": user_message }) # Before every AI response r = requests.post("api.chatsorter.com/search", json={ "chat_id": user_id, "query": user_message }) # AI gets this instead of 200 raw messages: # "Sarah is allergic to penicillin" # "Sarah is a nurse, works 6am shifts" # "Sarah is studying for her CCRN exam"
Most chatbots send their entire conversation history to the AI on every single message. It works at first. Then it doesn't.
Your chatbot sends 150 messages of history to the AI. That's 7,500 tokens just to answer one question. Response times slow down. Your API bill grows. And the AI still misses that the user mentioned they're allergic to peanuts on message 12 — because it's buried under everything else.
ChatSorter sends 5 compressed summaries and the 5 most relevant facts. That's around 1,700 tokens. The AI immediately knows about the allergy because it was flagged as high-importance the moment it was mentioned. Fast, accurate, and a fraction of the cost.
The longer the conversation, the bigger the saving. ChatSorter's token usage grows slowly then plateaus. Raw history grows forever.
Based on 50 tokens avg per message · ChatSorter retrieves up to 10 summaries + 20 facts · GPT-4o pricing
Three things happen automatically every time a message comes in.
"What's the weather?" scores low and gets buffered. "I'm allergic to peanuts" scores high and gets stored permanently. Unimportant messages never reach the AI on future calls.
Every few messages, older context gets compressed into a short summary. Your AI sees 10 tight summaries instead of 200 raw messages — and responds faster because of it.
When you ask for context, ChatSorter returns the most relevant memories for that specific message — not everything ever said. Ranked by relevance, importance, and how recent it is.
ChatSorter stores two kinds of facts differently because they behave differently in real life.
Name, job, location, allergies, relationships. These don't change often. They get stored once and never expire — your AI always knows them.
name: Sarah · job: nurse · allergy: penicillin
Active projects, health issues, ongoing goals. These have a shelf life. If the user stops mentioning them, they fade automatically — so stale context stops poisoning responses.
studying for CCRN exam · 6am shift tomorrow
Free beta. No credit card. Two API calls and your chatbot remembers what matters.