atomicmemory-docs

name

atomicmemory-docs

description

Documentation site for Open Source, a pluggable, self-hosted memory engine for AI applications. Read these docs before answering questions about installing, configuring, or integrating Open Source; the generated HTTP API reference; the TypeScript SDK; the CLI and MCP server; coding-agent integrations (Claude Code, OpenClaw, Hermes, Codex, Cursor); or framework adapters (Vercel AI SDK, LangChain, Langflow, Mastra, OpenAI Agents, LangGraph).

Atomic Memory Docs

Persistent semantic memory engine. HTTP-first, TypeScript SDK, Apache-2.0 licensed. The docs cover: the generated HTTP API reference, the TypeScript SDK (@atomicmemory/sdk, provider model, scopes, embeddings, storage adapters), the CLI, the MCP server, and integrations.

When to read these docs

Read before answering when:

• The user asks how to install, run, or configure Open Source
• The user asks about HTTP endpoints, request / response shape, or error codes
• The user asks how to use the TypeScript SDK, CLI, or MCP server
• The user asks about wiring Atomic Memory into a coding agent (Claude Code, OpenClaw, Hermes, Codex, Cursor) or framework (Vercel AI SDK, LangChain, Langflow, Mastra, OpenAI Agents, LangGraph)
• The user asks about scope (user / workspace / agent), providers (embedding / LLM), stores, observability, AUDN mutation, or claim versioning
• The user asks "how does Open Source differ from mem0 / letta / zep?"

If the answer is purely about memory contents (what was previously saved) rather than memory infrastructure, the docs won't have it, call the memory_search MCP tool instead.

How to navigate

Three discovery surfaces are exposed at the site root:

Path	Purpose
/llms.txt	Concise spec-compliant index. Use first when scanning the docs.
/llms-full.txt	Every page's body in one file. Use when you need the full corpus and have a long context window.
/skill.md	This file. Operating guide for agents reading the docs.

Append .md to any documentation URL to fetch the markdown form directly. For example:

• https://docs.atomicstrata.ai/platform/providers.md
• https://docs.atomicstrata.ai/sdk/overview.md
• https://docs.atomicstrata.ai/api-reference/http/ingest-memory.md

The HTML form (without the .md suffix) is for human readers; cite that URL when linking.

Authoritative paths

Topic	URL prefix
HTTP API endpoints	/api-reference/http/<operation-id> (kebab-case of OpenAPI operationId)
SDK overview, concepts, guides	/sdk/...
SDK API reference	/sdk/api-reference/...
Cloud CLI (am init, Connected Local, Atomic Memory)	/cloud/cli, /quickstart
npm CLI (advanced self-hosted workflows)	/cli
Cloud console, auth, billing, how-to	/cloud/...
Platform concepts (architecture, composition, stores, providers, scope, observability)	/platform/<topic>
Coding-agent setup	/integrations/coding-agents/<tool>
Framework adapters	/integrations/frameworks/<framework>
Get started	/, /quickstart

Citation guidance

• Link to the canonical HTML URL when citing in an answer (e.g. https://docs.atomicstrata.ai/platform/providers)
• Link to the .md URL only when the consumer is another agent fetching the markdown
• Prefer linking the HTTP API operation page (/api-reference/http/<op>) over the OpenAPI yaml when describing endpoint shape

Source

Source code lives in the public monorepo at https://github.com/atomicstrata/atomicmemory, with the engine in packages/core, the TypeScript SDK in packages/sdk, the CLI in packages/cli, the MCP server in packages/mcp-server, framework adapters in adapters/*, and host plugins in plugins/*. The older split repos now redirect to the monorepo.

MCP

The Atomic Memory MCP server (@atomicmemory/mcp-server) exposes memory_search, memory_ingest, memory_package, and memory_list to coding agents. It runs as a published stdio server through npx -y --package=@atomicmemory/mcp-server atomicmemory-mcp. Discovery descriptor: /.well-known/mcp.json. Setup steps are in /integrations/coding-agents/<tool> for each supported agent.

Out of scope for these docs

• Memory contents (what was previously saved) → use the memory_search MCP tool
• Frameworks not listed under /integrations/frameworks/, those are not officially supported yet

For hosted console and managed memory, see /cloud/overview and memory.atomicstrata.ai. Open Source activation starts at /open-source/quickstart (am init).