AI Companion With Memory

HoneyChat uses a two-tier memory system. Short-term: last 20 messages stored in Redis with 7-day TTL. Long-term: ChromaDB vector embeddings store semantic meaning of past conversations. Top-K relevant memories are retrieved and injected into each response context.

Long-term memory persists indefinitely. ChromaDB stores semantic embeddings of all conversations — emotional moments, key facts, relationship milestones. Short-term context covers the last 20 messages. Old history is automatically summarized to fit within token limits.

Names, preferences, shared stories, emotional context, relationship dynamics, inside jokes, and key events. The semantic memory retrieves what is relevant to the current conversation — not a random dump of old messages. Memory improves the more you chat.

Basic memory works on all plans including free. Paid plans increase the context window (how much memory is loaded per response) and use higher-quality AI models that better utilize memory context. Elite tier gets the largest context and best model.

Yes. You can request memory deletion at any time. Chat history can be cleared, and ChromaDB embeddings removed. This effectively resets the relationship with your companion to a fresh start.

ChatGPT memory stores explicit facts you tell it. HoneyChat uses semantic vector memory — it captures emotional nuance, relationship dynamics, and narrative context automatically. Plus it works inside Telegram with character voices and photo generation.

Yes. All characters, including custom-created ones, use the same ChromaDB semantic memory system. Your custom character will remember your interactions, develop relationship dynamics, and reference past conversations just like pre-built characters.

Yes. Memory is tied to your Telegram chat and not shared with anyone. No personal data is collected beyond your Telegram user ID. ChromaDB data is stored on encrypted infrastructure and can be deleted on request.