Agents Memory

Agents Memory is a local-first memory runtime for AI agents. It organizes project rules, session checkpoints, active tasks, long-term memory, and sensitive secret references into a searchable context system. It is built for personal development environments where Codex, Claude, Cursor, or other agents need to resume work without repeating project background after context compaction.

The current version provides the am CLI, project registration, session start, checkpoints, context retrieval, active indexes, legacy memory migration, index rebuilding, and am-secrets secret reference management. The new get-context --profile lean|normal|deep option lets agents choose a context pack size: lean for quick handoff, normal for daily work, and deep for complex tasks that need more history.

Agents Memory does not rewrite user-owned AGENTS.md, CLAUDE.md, or .cursorrules; it backs up existing files and appends only a managed block. Sensitive values should stay in AM secrets, while ordinary memory and docs keep only secret_ref references.