@atomicmemory/cli

@atomicmemory/cli

Human- and agent-facing CLI for AtomicMemory memory workflows.

This package is separate from @atomicmemory/mcp-server: atomicmemory-mcp is a stdio protocol process for agent hosts, while atomicmemory is a normal command-line tool.

Install

From a local checkout, build and link globally:

pnpm --filter @atomicmemory/cli build
cd packages/cli
pnpm link --global

atomicmemory

With no arguments in a real terminal, atomicmemory opens the interactive Ink UI. Use atomicmemory help for the plain command reference, or atomicmemory --no-interactive to keep the static help behavior.

If pnpm reports ERR_PNPM_NO_GLOBAL_BIN_DIR, initialize pnpm's global bin directory first, then restart or source your shell config:

pnpm setup
source ~/.zshrc
cd packages/cli
pnpm link --global

You can also test without global linking:

node packages/cli/dist/bin.js

The interactive UI can also be opened explicitly:

atomicmemory ui
atomicmemory --interactive

Public npm install will be:

npm install -g @atomicmemory/cli

That requires publishing @atomicmemory/sdk first. Until the SDK exists on npm, this package keeps the same local SDK file: dependency used by the rest of the integrations repo.

Usage

atomicmemory
atomicmemory init --api-url http://127.0.0.1:17350 --user "$USER"
atomicmemory doctor
atomicmemory status
atomicmemory add "The project uses pnpm workspaces."
atomicmemory search "workspace package conventions" --limit 5
atomicmemory package "recent implementation context" --token-budget 1200
atomicmemory list --limit 20
atomicmemory get <memory-id>
atomicmemory delete <memory-id>
atomicmemory skill get core
atomicmemory hooks install --host codex --runtime node
atomicmemory validate

Every command accepts the same provider and scope overrides:

atomicmemory search "release policy" \
  --provider atomicmemory \
  --api-url http://127.0.0.1:17350 \
  --user "$USER" \
  --namespace atomicmemory

The legacy atomicmemory memory <command> group remains as a compatibility alias, but the v4 command spec treats the direct commands as canonical:

atomicmemory search "release policy"
atomicmemory add "Remember this fact"
atomicmemory list

Command Groups

The visible v5 surface is driven by cli-spec.json:

Group	Commands
setup	init, config, hooks, completion
diagnose	doctor, status, validate
agent	skill, help, version
memory	add, ingest, search, package, list, get, delete, import

Use atomicmemory help --json for the machine-readable command tree.

Hook runtime selection

atomicmemory hooks install emits host-specific lifecycle hook config without mutating user config files. Node is the recommended default and is bundled as atomicmemory hooks run .... Python is an advanced option for teams that set ATOMICMEMORY_PYTHON_HOOK_BIN to a compatible Python hook runner.

atomicmemory hooks install --host codex --runtime node
atomicmemory hooks install --host codex --runtime python
atomicmemory hooks install --host claude-code --runtime node

Agent output

Use --agent for stable JSON envelopes:

atomicmemory memory search "prior decisions" --agent

Success shape:

{
  "status": "success",
  "command": "search",
  "duration_ms": 12,
  "profile": "default",
  "scope": { "user": "pip" },
  "count": 1,
  "data": {}
}

Errors in agent mode are emitted to stdout and exit non-zero.

Config

The default config path is ~/.atomicmemory/config.json.

Precedence is:

• CLI flags
• ATOMICMEMORY_* environment variables
• local config file
• command defaults

Memory commands require an explicit scope user from flags, environment, or config. The CLI does not invent a user for provider-backed operations.

atomicmemory config show redacts API keys by default. Use atomicmemory init --api-key-stdin to avoid putting a token in shell history.

Backend smoke (Docker)

The backend-gated smoke suite (pnpm -C packages/cli test:backend) is skipped unless ATOMICMEMORY_TEST_BACKEND=1 is set against a real @atomicmemory/core instance. To exercise it deterministically against a local Docker stack, run:

pnpm -C packages/cli test:backend:docker

The harness (scripts/test-backend-docker.mjs) brings up docker-compose.yml + docker-compose.smoke.yml from packages/core, layers a small CLI-side overlay (scripts/docker/docker-compose.cli-backend.yml) that routes core's LLM at an in-network mock so /v1/memories/ingest doesn't 401 on core's smoke OPENAI_API_KEY=sk-smoke-test-dummy, polls the real /health endpoint with a bounded 90-second timeout (no sleep-only readiness), runs pnpm build + the backend suite against the resolved API URL, then tears the stack down. No external API credentials are required.

Configuration env vars (all optional):

Var	Default	Purpose
ATOMICMEMORY_CORE_PATH	sibling ../core	path to the core package root
ATOMICMEMORY_DOCKER_APP_PORT	first free port from 3060	host port for core's app
ATOMICMEMORY_DOCKER_POSTGRES_PORT	first free port from 5444	host port for core's pg
ATOMICMEMORY_DOCKER_HEALTH_TIMEOUT	90 (seconds)	bounded /health poll cap
ATOMICMEMORY_DOCKER_HEALTH_INTERVAL	2 (seconds)	poll interval
ATOMICMEMORY_DOCKER_SKIP_BUILD	0	reuse existing compose images
ATOMICMEMORY_DOCKER_KEEP_UP	0	leave the stack up after the run for inspection

Requirements: docker daemon running, docker compose v2 plugin, pnpm, and packages/core containing both docker-compose.yml and docker-compose.smoke.yml.