Launching today
Belcore
Give AI agents memory across sessions
1 follower
Give AI agents memory across sessions
1 follower
Belcore sits in front of your LLM and keeps what matters from a conversation, so the model gets a compact, relevant context instead of the full history. Works with any chat stack. Try the live demo, no sign-up.
Launch Team
AWS Activate CreditsCredits that fuel your startup’s next milestone
Promoted
Hi everyone, I'm the maker of Belcore.
Chat and agent apps usually resend the whole history on every call, so input tokens grow with the session. Belcore keeps a memory of what was said and sends the model a selected context instead.
On LongMemEval_S (500 questions, about 109k tokens of prior conversation each), the assembled memory context was 6,671 tokens versus 109,079 for the full transcript: 93.9% fewer input context tokens in this benchmark, compared with sending the full transcript. This is context size only. Billed input per call is higher because the prompt and question are added, and accuracy and cost per successful task are measured separately.
You can try it on the landing page without signing up. The panel next to the chat shows what it remembers from your messages. It is a limited public demo.
I'd like to hear where it breaks, especially for long sessions and multi-session recall.