rightmemory

RightMemory gives AI coding agents three semantic modules: durable Memory for context that persists beyond individual sessions, a Pursuit map representing the user's active directions hierarchically, and Agent Corrections storing reusable cases of user redirection. Memory and Pursuit are Markdown document trees within one addressable graph, while Agent Corrections remains a fixed, non-graph case library. The system addresses a common problem with coding agents: they perform well within a single conversation but lose context afterward. RightMemory treats memory as owned project and collaboration state, supporting multi-agent collaboration through shared views where roots can share selected context without exposing the entire private store. Key capabilities include multi-device continuity so the same memory follows agents across laptops, desktops, and different agent clients; clear ownership boundaries so retrieval, updates, and consolidation run through explicit roles rather than letting agents half-edit their own memory; and vendor-neutral access via a `rightmemory` CLI or JSON-over-stdio daemon surface compatible with Codex SDK, Claude Code CLI, and other command-capable agents. Memory and Pursuit use separate lifecycles inside one graph with typed edges (dep:, cfg:, ver:, doc:, todo:) and a globally unique id namespace. Headings support anchors, F# detail file references, and M#/S# linked resources. The system includes automatic orchestration plus independent maintenance skills for map management, Memory upkeep, guidance review, and a Manager mode for long-lived coordination conversations. The memory root is a standalone Git working tree containing MEMORY.md, PURSUITS.md, and supporting files. Named profiles allow project-specific memory roots. Shared views connect one root to external collaboration context through MF# (mirrored files) and MQ# (provider questions) heading types. A Web Studio provides a visual pursuit map editor, and a background watcher handles pruning, Dreamer consolidation, and Insight reflections. Installation uses uv and git, with scripts for macOS, Linux, and Windows. Skills install into ~/.codex/skills and ~/.claude/skills by default. The system avoids vector databases and raw notes in favor of deterministic Markdown with inspectable structure and explicit graph relationships.