atomicstrata/atomicmemory

npm install @atomicmemory/sdk
npm install -g @atomicmemory/cli
npm install @atomicmemory/vercel-ai @atomicmemory/sdk (framework adapter example)
Core backend is Docker-deployable (packages/core Dockerfile)

Usage

• •Create a MemoryClient with a providers config (e.g. apiUrl http://localhost:17350), then memory.initialize(), memory.ingest({ mode: 'messages', ... }), and memory.search({ query, scope })
• •Use the CLI for setup, diagnostics, capture, retrieval, and context packaging
• •MCP server (@atomicmemory/mcp-server) and host plugins (Claude Code, OpenClaw, Hermes) install per host contract

Key features

• ✓Portable single memory protocol consumed by SDK, CLI, MCP server, framework adapters, and host plugins
• ✓Inspectable Core: open source, self-hostable, durable context with Postgres/pgvector storage
• ✓Correction-aware memory: supersession, clarification, deletion, no-op decisions, lineage, and trust-sensitive revision
• ✓Framework adapters for Vercel AI SDK, OpenAI Agents SDK, LangChain, LangGraph, Mastra
• ✓Host plugins for Claude Code, OpenClaw, Hermes (Codex and Cursor plugin source present but 'coming soon')
• ✓Benchmark claims: leading performance/cost on BEAM-100K, BEAM-1M, LoCoMo10 under matched methodology

Best for

Teams that want self-hostable, inspectable agent memory with a portable, model-agnostic interface

Caveats

• ⚠Codex and Cursor host plugins are 'coming soon' — not installable until marketplace manifests are validated
• ⚠Python SDK lives in a separate repository (not part of this monorepo)
• ⚠Until latency benchmarks are linked, treat the engine as 'designed for single-digit-ms local retrieval' — a design target, not a guarantee
• ⚠Core DB-backed tests require Postgres/pgvector provisioning