A persistent knowledge graph for AI coding sessions that solves the cold start problem. Exposes six MCP tools: suma_ingest writes architectural decisions and bug fixes to a weighted graph, suma_search retrieves context by natural language query, suma_talk combines search and learning in one call, and suma_correct supersedes wrong information without deletion. Runs as a hosted service on Cloud Run, so no local server setup. The K-WIL gravity algorithm ranks facts by recency, density, semantic similarity, and emotional weight rather than flat chunking. Reach for this when you're tired of re-explaining your auth flow or database schema every time you open a new Claude chat, or when multiple agents need to share context across sessions without explicit handoffs.
claude mcp add --transport stdio io.github.sumanaddanki-suma-memory uvx suma-memory