Memory infrastructure
that doesn't cost a fortune.

Mem0, Supermemory, and Letta run knowledge graphs on cloud compute. ChatSorter runs on your machine or a $5 VPS. Same persistent memory. A fraction of the cost.

Mem0 ~$0.002/op vs ChatSorter $0.00/op
·
Supermemory cloud compute vs ChatSorter your hardware
·
Letta knowledge graphs vs ChatSorter local LLM
Free Beta
Hosted on ChatSorter servers. Try everything, no commitment.
$0 /forever
No credit card required
  • 50,000 API calls / month
  • All three memory layers
  • Importance scoring
  • Fact extraction + summaries
  • Dashboard access
  • Community support
  • Data stored on ChatSorter servers
Start free →
Coming Q3 2026
Managed Pro
Hosted on ChatSorter with higher limits and priority support.
$49 /month
500,000 API calls/month
  • 500,000 API calls / month
  • All features included
  • Priority email support
  • 99.5% uptime SLA
  • Dashboard + analytics
  • Data stored on ChatSorter servers

Need a custom volume deal or enterprise compliance requirements? Email us.

Why cheaper isn't worse.

The expensive tools run knowledge graphs on GPU clusters. ChatSorter uses importance scoring + a small local LLM — simpler architecture, same persistent memory, dramatically lower cost.

Tool Architecture Per-operation cost Your API token savings Data location
Mem0 Knowledge graph + cloud LLM ~$0.002 per memory op Yes Their servers
Supermemory Vector DB + cloud compute Cloud pricing Partial Their servers
Letta Knowledge graphs + agents High compute cost Yes Their servers
ChatSorter Local LLM + importance scoring $0.00 / op (self-hosted) Up to 93% fewer tokens Your server

Common questions.

How does self-hosted actually work?

You download a single executable, add your license key to a config file, and run it. It starts a local API server on your machine. Your chatbot calls localhost instead of our servers. Conversation data never leaves your infrastructure — the only thing that contacts our server is a daily license check (just the key, no data).

What hardware do I need for self-hosted?

Anything that can run Python and Ollama. A $5/month VPS handles low-to-medium traffic fine. A standard developer laptop handles testing and development. The default model (qwen2.5:3b) runs on CPU if you don't have a GPU, though a GPU makes summarization faster.

What's the "93% token reduction" based on?

At message 500, a raw conversation history sent to your LLM is ~25,000 tokens. ChatSorter returns up to 10 compressed summaries and 20 extracted facts — about 1,850 tokens. That's a 93% reduction in input tokens on every request. The longer the session, the bigger the saving. Full calculation on the homepage.

Can I switch from free beta to self-hosted?

Yes. Purchase a self-hosted license, download the executable, and point your integration at localhost instead of api.chatsorter.com. The API is identical — same two endpoints, same JSON format.

What happens if my license expires?

The local server will stop accepting requests on the next restart after expiry. You'll get a clear message with a link to renew. Renewing restores full access immediately.

Is there a free trial for self-hosted?

The free beta hosted version lets you test the full API with no time limit. When you're ready to move data off our servers, upgrade to self-hosted. Email us if you want to discuss a trial period.

The budget-conscious memory layer.

Start free. Move to self-hosted when your data needs to stay yours.

Try the demo Get started free →