- 01 Your applicationExisting LLM client & agent tools
- 02 MemoriCapture, recall & augment
- 03 SQL memoryContext that persists across interactions
Useful context. Familiar infrastructure.
The problem
AI applications need useful context across interactions. Adding memory should fit the application’s existing clients, data, and infrastructure while keeping irrelevant context out of the prompt.
My contribution
I own core engineering for the async-first Python SDK and memory platform. My work spans retrieval pipelines, provider integrations, MCP and Claude Code workflows, and production AWS infrastructure.
A key design choice
Combine embeddings, vector search, graph context, and reranking to select useful memories before augmentation. A pluggable provider architecture lets integrations fit different LLM ecosystems.
The result
An open-source SDK and memory platform with documented agent integrations and production deployment paths. I also own release quality through automated tests, type checks, CI, and the infrastructure behind Memori Cloud.