Every message you send gets scored, facts get extracted, and summaries get generated. Watch the memory panel on the right update in real time.
POST /process. the same endpoint your chatbot would call. The importance score and extracted facts came back in the response JSON.
messages_per_summary. Short conversations compress faster; long evolving threads (like a to-do list) benefit from a higher value so items don't get split across summaries.
POST /search before each LLM call. You get back the top memories ranked by similarity × importance × recency, ready to drop into your system prompt.
Ready to add this to your chatbot?
Get your free API key →