Kindex

A persistent knowledge graph that gives AI agents durable memory across sessions. It indexes conversations, projects, and decisions into a weighted graph with types, domains, and decay, then injects context at five budget tiers (100 to 4000 tokens) so you never waste the window. Exposes 30+ MCP tools including search, add, link, ask (with Anthropic caching), learn, constraints, directives, and session tagging. Ships with setup commands for Claude Code, Codex, Gemini CLI, OpenCode, and Cursor, plus instruction files that teach agents to actually use it. The graph understands relationships and surfacing, not just logs. Supports repo-level .kin configs with inheritance chains for team and org contexts. Reach for this when you're tired of agents forgetting project constraints or re-asking questions you already answered three sessions ago.