Integrations
Agent index: llms.txt
AtomicMemory adds durable memory to coding agents, AI frameworks, and terminal workflows through the same provider-backed memory layer. Choose the host you use, install the local integration, and point it at your AtomicMemory backend.
Coding agents
| Integration | Best for | Status | Docs |
|---|---|---|---|
| Claude Code | MCP tools, memory skill, lifecycle hooks, and local core connection. | Published; hosted mode planned. | Overview · Local |
| Codex | MCP tools plus a memory protocol skill for task-start recall and handoffs. | Manual; packaged plugin planned. | Overview · Local |
| OpenClaw | Published plugin with embedded MCP tools and a memory skill bundle for cross-channel agent memory. | Published; hosted mode planned. | Overview · Local |
| Hermes Agent | Published native Python memory provider with prefetch, turn sync, and explicit tools. | Packaged provider; hosted mode planned. | Overview · Local |
| Cursor | MCP tools plus Cursor rules for durable memory behavior. | Manual local MCP + rule template; packaged plugin planned. | Overview · Local |
Frameworks
| Integration | Best for | Status | Docs |
|---|---|---|---|
| Vercel AI SDK | Pre-call retrieval and post-call ingest around generateText / streamText. | Published; hosted mode planned. | Overview · Local |
| OpenAI Agents SDK | Memory-aware run() flows and optional function tools. | Published; hosted mode planned. | Overview · Local |
| LangChain (JS) | Memory search/ingest tools and helper functions around an injected SDK client. | Published; hosted mode planned. | Overview · Local |
| Langflow | Custom visual-flow components for explicit store, prompt-ready search context, read-only chat memory, and scoped deletion. | Published; hosted mode planned. | Overview · Local |
| LangGraph (JS) | Retrieve and ingest node factories for durable memory inside state graphs. | Published; hosted mode planned. | Overview · Local |
| Mastra | Memory search/ingest tools for Mastra agents. | Published; hosted mode planned. | Overview · Local |
Shared memory surface
Most coding-agent integrations use the shared MCP server from the
packages/mcp-server
package in the public
atomicstrata/atomicmemory
monorepo:
| Tool | Purpose |
|---|---|
memory_search | Search durable memory by meaning. |
memory_ingest | Store durable facts, decisions, preferences, or exact session snapshots. |
memory_package | Build a token-budgeted context package. |
memory_list | Inspect recent scoped memories. |
Framework integrations use the TypeScript or Python SDK directly, but keep the same loop: retrieve before the agent acts, ingest after useful work completes, and scope memory by user, agent, namespace, or thread.
Terminal users can use the AtomicMemory CLI for direct search, ingest, profile setup, and hook generation.
Backend choice
Integrations are backend-agnostic. The SDK's MemoryProvider model lets the same integration point at self-hosted atomicmemory-core or another registered provider by configuration.
Contributing
Source lives in the public
atomicstrata/atomicmemory
monorepo:
- MCP server:
packages/mcp-server - Framework adapters:
adapters/* - Host plugins:
plugins/*