Assistant memory
The assistant's memory of your conversations — facts you've shared, decisions you've made, and context from past chats. It's automatically extracted and retrieved when relevant, so the assistant gets smarter the more you use it.
GreatChat remembers context across conversations. It stores facts, preferences, and decisions — and retrieves them when relevant — so it gets smarter the more you use it.
The assistant's memory of your conversations — facts you've shared, decisions you've made, and context from past chats. It's automatically extracted and retrieved when relevant, so the assistant gets smarter the more you use it.
Facts you explicitly teach GreatChat — your role, your preferences, your tools, your goals. These persist across all conversations and shape how the assistant responds to you personally.
GreatChat remembers through a vector memory system: conversations are turned into embeddings, stored, and retrieved when relevant. It's not a giant prompt — it's semantic search over your past.
Facts you share are extracted from conversation — preferences, names, dates, decisions, and recurring instructions.
Text becomes vectors via an embedding model. This turns meaning into numbers so similar concepts are close together.
Vectors live in a vector database with COSINE similarity search. Your memories are isolated to your account — never shared.
When you ask a question, the best matching memories are pulled into context automatically. The assistant recalls what matters.
Memory isn't a static log you have to maintain by hand. It's a living system that organises, reinforces, and prunes itself as you keep talking — so the longer you use GreatChat, the more sharply it understands you.
You never file memories manually. As you talk, the system quietly pulls out durable facts — names, preferences, decisions, and recurring instructions — and stores them the moment they matter.
Memories are connected by meaning, not just keywords. A note saved weeks ago resurfaces in the right conversation even if the wording is completely different, because the model understands the relationship.
The more often a fact recurs, the stronger it becomes. Repeated preferences and decisions are promoted to long-term context, so the assistant leans on what it knows you truly want.
Stale and one-off details fade instead of piling up. Low-signal memories are demoted over time, so your context stays relevant and the assistant isn't cluttered with yesterday's throwaway context.
Memory carries across every chat, not just one thread. Start a new conversation next week and the assistant already knows the project, the people, and the plan — no re-explaining.
Memories are organised into types and topics — people, projects, preferences, facts — so retrieval is precise. The assistant grabs the exact slice of context it needs instead of a noisy dump.
The more you chat, the more context the assistant has. It remembers your preferences, past decisions, and project details — so you never repeat yourself.
Memory isn't keyword matching — it's meaning-based. Ask "what did we decide about the launch?" and it finds the right memory even if you used different words.
View, edit, and delete any memory at any time. Your memories are yours — isolated to your account and never used to train shared models.
Tell GreatChat about yourself and watch it get smarter with every conversation.
Start chatting →