Mira — conversational agent with long-term memory

Problem

Conversational agents forget everything between sessions. Anything worth remembering has to be re-supplied by the user, which caps how useful the agent can be over time.

Approach

Memory layer built end-to-end on a structured Markdown knowledge base: a defined note schema, automated extraction and writing of salient information, and retrieval of relevant prior context at inference time. Conversation-state handling keeps continuity and tone coherent across extended multi-session use.

Stack

• Python
• LLM APIs
• retrieval / RAG
• Markdown knowledge base