Agent Memory System

Agent Memory System is an AI developer tool that provides a durable, secure memory layer for every repository in a workspace, giving AI agents persistent project context across sessions for software developers and engineering teams. Built by RAVBYTE TECHNOLOGIES PRIVATE LIMITED and available under the MIT license, the tool installs with a single npx command and generates structured memory files including context-index.json, agent-guidelines.md, agent-worklog.jsonl, and agent-handoff.md.

The repository scanner maps every folder and repo in a workspace in one command. It detects manifests, routes, APIs, configs, tests, storage hints, documentation, and generated directories, then writes Markdown plus a topic index that any supported agent can read. This eliminates the cold-start problem where each new AI session has to re-learn the structure of a codebase from scratch.

The agent-memory maintain --since main command detects structural Git changes, refreshes the memory directory, and validates the result so stale context does not quietly accumulate between reviews. Built-in static analysis generates a dependency graph with O(1) query time, letting agents determine the blast radius of any change before making it — without guessing which files share an import chain.

Agent Memory System is designed to work with Codex, Claude, Cursor, Antigravity, and future agents through a portable skill wrapper. The cross-agent handoff mechanism records checkpoints, commands, files touched, blockers, and next steps in a JSONL worklog, then generates a handoff Markdown file so the next assistant can resume without losing working state. Security is built in: environment variable names are recorded without exposing .env values, JWT-like strings and long hex values are flagged, and generated paths such as node_modules and dist are excluded from source ownership records.

Agent Memory System is used by software developers and engineering teams to give AI coding agents persistent repository context, enabling cross-session continuity and structured agent handoffs. It is particularly suited for teams using multiple AI agents (Claude, Codex, Cursor) on the same codebase who need to eliminate repeated cold-start context loss.