AtomicMemory

Inspectable, correction-aware memory for agents and AI applications.

AtomicMemory gives agents durable context across sessions without coupling your application to one model, framework, or deployment. Start with managed Hosted Cloud, run the open-source Core locally, or integrate through the TypeScript SDK and MCP server.

Documentation · Hosted Cloud · Open-source quickstart · Why inspectable memory matters

Headline Results

AtomicMemory v66 is leading performance/cost on BEAM-100K, BEAM-1M, and LoCoMo10 under matched methodology against published competitors. On BEAM-10M it matches the strongest published Mem0-new result while leaving Hindsight-scale temporal retrieval as the known open frontier.

Benchmark	AtomicMemory v66	Position	Cost/Q	Sample
BEAM-100K lenient	0.7375	Parity with Hindsight at 0.75	$1.26	n=80
BEAM-1M lenient	0.6625	Leading Performance/Cost; +0.022 vs Mem0 paper	$0.083	n=80
BEAM-10M lenient	0.4875	Parity with Mem0-new at 0.486	$0.081	n=80
LoCoMo10 GPT-4o-mini binary	0.8396	Leading Performance/Cost; +0.171 vs Mem0 paper	$0.066	n=1540

These results put AtomicMemory at or near the published ceiling in each reported category while preserving the lower-cost operating profile that matters for real applications. Reproducibility artifacts and harness details will be published with the benchmark materials.

Quickstart

Install the am CLI and initialize Hosted Cloud in one guided command:

curl --proto '=https' --tlsv1.2 -fsSL https://get.atomicstrata.ai/install.sh | sh -s -- --init

Plain am init defaults to managed Cloud, selects your project, and saves its project-bound credential. Managed server keys use a per-installation name such as am-cli-a1b2c3d4e5f6, so another machine's credential is not rotated. No Docker or OpenAI key is required. Non-interactive Cloud automation uses am init --yes --project <cloud-id>; Local automation must opt in with am init --local --yes.

The installer runs initialization through the verified binary directly. If it updates PATH, open a new terminal before the next commands or use the shell-specific activation command it prints.

Store a preference and retrieve it:

am memory ingest "I prefer aisle seats when flying."
am memory search "seat preference"

Connect the active profile to an agent host when you are ready:

am integrate --yes --host cursor # or claude-code / codex

am integrate writes the host's user-level MCP configuration. It does not install a marketplace plugin.

Choose your path

Path	Best for	Start here
Hosted Cloud	Managed memory with the fastest setup	Guided installer above, or am init
Connected Local	Running open-source Core on your machine	am init --local
TypeScript SDK	Server-side application integration against Core	npm install @atomicmemory/sdk

Agent integration through MCP works with either an active Cloud or Local profile. See the documentation for framework and host-specific guides.

Why AtomicMemory

• Correction-aware — supersede, clarify, delete, or retain memories as facts change instead of treating memory as append-only recall.
• Portable — use one memory protocol through the CLI, MCP server, SDK, framework adapters, and host plugins.
• Inspectable — run the open-source Core and audit the mutation and retrieval path rather than depending only on a hosted black box.
• Model-flexible — keep extraction, embeddings, mutation, reranking, and retrieval packaging behind explicit provider boundaries.
• Cloud or Local — begin with managed Cloud or operate Core yourself without rewriting the integration surface.

Use AtomicMemory

Hosted Cloud

Interactive am init offers Hosted Cloud as option 1/default and Connected Local as option 2. Hosted Cloud needs no Docker or OpenAI key:

• With one project, the CLI selects it automatically. With multiple projects, it prompts for a selection.
• With no project, it opens onboarding and waits for project creation in an interactive terminal. Non-interactive runs print the URL and recovery command instead of waiting.
• Credentials are bound to the Cloud origin and project, stored with owner-only permissions, and never printed.
• A working stored project credential is reused. Otherwise the CLI rotates only this installation's exact am-cli-<12-hex> key and creates it when absent. Legacy unsuffixed keys and other installations' keys are left untouched.

--project accepts a unique ID or case-insensitive slug and infers Cloud versus Local from the resolved project type. Explicit selectors assert the type:

am init --project <id-or-slug>          # infer Cloud or Local
am init --cloud --project <cloud-id>    # require a Cloud project
am init --local --project <local-id>    # require a Local project

Ambiguous slugs fail with a request for the unique project ID. At the API-key limit, initialization preserves the previous default profile and prints the dashboard URL plus exact commands to list or revoke a key and retry. It never rotates or revokes unrelated keys as quota recovery.

Connected Local

Connected Local runs Core on your machine and can link it to Cloud for trace visibility. It requires:

• Docker Desktop or Docker Engine running
• An OpenAI API key; interactive setup reads it with hidden input and stores it with owner-only permissions
• macOS or glibc Linux on x86_64 or arm64

Initialize and verify Local:

am init --local
am doctor --smoke

The Local defaults remain profile local and Core URL http://127.0.0.1:17350. For headless automation, seed dashboard auth and the OpenAI key before selecting Local explicitly:

am auth login --token "$AM_DASHBOARD_JWT"
export OPENAI_API_KEY=sk-... # inject through your secret manager
am init --local --yes

For Core-only Docker without a Cloud account, use the Core package guide. The open-source quickstart covers lifecycle, custom URLs, and troubleshooting; the CLI README documents every initialization flag.

Agent hosts and MCP

Both initialization paths leave an active profile that the published MCP server can use. Configure a supported host with:

am integrate --yes --host cursor # or claude-code / codex

Codex and Cursor marketplace plugin packages remain coming soon in the matrix below; direct MCP configuration through am integrate is a separate supported path. Use am integrate doctor to diagnose host configuration.

TypeScript SDK

The SDK is server-side only in v1. Start Core first, then reveal the Local client environment only in a trusted terminal:

am connect env --for clients --show-secrets
npm install @atomicmemory/sdk

Copy ATOMICMEMORY_CORE_URL and CORE_API_KEY into the trusted server process; never expose CORE_API_KEY in a browser bundle. Then use the memory client:

import { MemoryClient } from '@atomicmemory/sdk';

const memory = new MemoryClient({
  providers: {
    atomicmemory: {
      apiUrl: process.env.ATOMICMEMORY_CORE_URL!,
      apiKey: process.env.CORE_API_KEY!,
    },
  },
});

await memory.initialize();
await memory.ingest({
  mode: 'messages',
  messages: [{ role: 'user', content: 'I prefer aisle seats.' }],
  scope: { user: 'demo-user' },
});

const results = await memory.search({
  query: 'seat preference',
  scope: { user: 'demo-user' },
});

Use AtomicMemoryClient when the application also needs the storage namespace. See the SDK quickstart and packages/sdk/README.md for the full API.

Installer verification

Every CLI download is checked against SHA256SUMS. If an authenticated GitHub CLI is available, the installer also verifies build provenance. Without GitHub authentication it warns, skips optional attestation, and continues only after checksum verification. To require attestation:

curl --proto '=https' --tlsv1.2 -fsSL https://get.atomicstrata.ai/install.sh | AM_VERIFY_ATTESTATION=1 sh -s -- --init

Required verification fails before installation unless gh auth login or GH_TOKEN supplies working GitHub authentication.

Package matrix

Status labels are part of the public docs contract:

• published — available on the npm registry and supported.
• implemented, publish pending — code lives in this repo and works locally, but the monorepo-era package has not been released.
• coming soon — source is present, but the public host install path is not supported yet.
• deprecated — still published and supported for the workflows named in its row, but superseded.

包

Package	Path	Status
@atomicmemory/core	packages/core	published
@atomicmemory/sdk	packages/sdk	published
@atomicmemory/cli	packages/cli	deprecated (published; use am, still required for llmwiki import)
@atomicmemory/mcp-server	packages/mcp-server	published
@atomicmemory/llmwiki	packages/llmwiki	implemented, publish pending

Framework adapters

Package	Path	Status
@atomicmemory/vercel-ai	adapters/vercel-ai	published
@atomicmemory/openai-agents	adapters/openai-agents	published
@atomicmemory/langchain	adapters/langchain	published
@atomicmemory/langgraph	adapters/langgraph	published
@atomicmemory/mastra	adapters/mastra	published

Host plugins

Package	Path	Status
@atomicmemory/claude-code-plugin	plugins/claude-code	published
@atomicmemory/openclaw-plugin	plugins/openclaw	published
@atomicmemory/hermes-plugin	plugins/hermes	published
@atomicmemory/codex-plugin	plugins/codex	coming soon
@atomicmemory/cursor-plugin	plugins/cursor	coming soon

Codex and Cursor plugin source is present, but the public host install path is coming soon until each host marketplace manifest format is validated end to end.

Other surfaces

Surface	Location	Status
CLI (am)	crates/cli	published; canonical artifacts on GitHub 发布 (get.atomicstrata.ai mirrors them)
Python SDK (atomicmemory on PyPI)	separate repository	published; not part of this monorepo

仓库 and trust boundaries

This repository is the public source of truth for AtomicMemory's JavaScript and TypeScript packages, Rust CLI, framework adapters, host plugins, and public smoke contracts. It contains:

• Core — Docker-deployable memory backend with mutation, retrieval, and Postgres/pgvector storage.
• SDK — typed provider boundary, memory and storage clients, embeddings, and search primitives.
• CLI and MCP server — setup, diagnostics, capture, retrieval, and agent integration surfaces.
• Adapters and plugins — thin integrations for supported frameworks and agent hosts.

Hosted service infrastructure, release orchestration, marketplace operations, the Python SDK, and unpublished benchmark harnesses live outside this monorepo. Package-specific setup remains in each package README.

Performance posture

The Headline Results above are benchmark scores under matched methodology. Latency, recall@k, and scale-envelope claims should only be quoted with the benchmark, hardware, dataset, and measurement date. Until linked latency benchmarks are available, single-digit-millisecond local retrieval remains a design target rather than a guarantee.

Public validation

拉取请求 verify repository hygiene, package metadata, affected build and type checks, lint, self-contained tests, package tarball shape, documentation contracts, public integration smoke, and security compliance. Package and release contexts additionally cover Core OpenAPI drift, schema tests, Docker image smoke, and DB-backed Core tests when their required services are present.

Local development

The monorepo uses pnpm workspaces and Turborepo. Check package.json for the complete command surface.

pnpm install
pnpm run build
pnpm run typecheck
pnpm run test
pnpm run lint

Core DB-backed tests require Postgres/pgvector provisioning. Rust CLI changes use the pinned toolchain and the repository's CI-equivalent command:

pnpm run ci:rust

See CONTRIBUTING.md for the full workflow, required checks, and package-level commands. The canonical CLI source and contributor setup are documented in crates/cli/README.md.

Companion: llmwiki

llmwiki compiles raw sources into an interlinked Markdown knowledge base. The @atomicmemory/llmwiki bridge imports its JSON export into AtomicMemory while preserving advisory metadata under memory.metadata.llmwiki.*.

See packages/llmwiki/README.md and packages/llmwiki/docs/cookbook.md for the full workflow.

Project information

• Release notes: package changelogs and CHANGELOG.md
• Roadmap: ROADMAP.md
• Contributing: CONTRIBUTING.md
• 安全: confidential reporting and supported versions in SECURITY.md
• License: Apache License 2.0 — see LICENSE

仓库 layout

packages/      core, sdk, cli, mcp-server
crates/        CLI (am) and Cloud/Core wire types
adapters/      framework integrations (Vercel AI, OpenAI Agents, LangChain,
               LangGraph, Mastra)
plugins/       host integrations (Claude Code, OpenClaw, Hermes, Codex, Cursor)
examples/      reserved for phase 2+; only added with owners and CI coverage
tests/smoke/   public, contributor-safe smoke tests

Release orchestration, marketplace operations, sensitive service configuration, and local machine paths are deliberately not part of this repository.