mopanc/saga

Saga is a local-first persistent memory layer for AI coding agents (Claude Code, Cursor, Windsurf, any MCP client). It stores curated, self-contained Markdown "topic" notes (~500 words) tied to code investigations so agents can recall context across sessions without re-explaining a codebase. Who uses it: solo devs and small teams embedding agent memory into workflows and IDEs. Key features: - Topic-grained notes with typed relations (@supersedes, @refines, etc.) - Layered scopes: personal, project, dept, org (personal syncs across machines; project lives in repo) - MCP JSON-RPC tools: recall, topic_read/list/write, plus auto-injection hook for Claude Code - SQLite full-text index + BM25 ranking; notes stored as versioned Markdown in git - Single static binary, cross-platform, no telemetry, secret detection and sync controls, open Saga Topic Spec.

AI-related keywords found in this repo's description, topics, or README summary — grouped by category. Each badge links to the corresponding ranking detail page.