Skip to content

quickstart.md

Latest commit

 

History

History
106 lines (71 loc) · 4.08 KB

quickstart.md

File metadata and controls

106 lines (71 loc) · 4.08 KB
Raw

Quickstart: give an assistant memory

Agent index: llms.txt

Public CLI release timing

This quickstart assumes a cloud-first CLI where plain am init sets up Hosted Cloud. If am init --help does not list --local, the public installer still serves an older flow — use plain am init for Connected Local instead and see Open Source Quickstart.

Outcome: store one travel preference, retrieve it in a later request, and confirm both operations in the console.

You need: macOS or Linux and an account at memory.atomicstrata.ai. No Docker required for the default Hosted Cloud path.

  1. 1Install and initializeRun the installer; plain am init sets up Hosted Cloud.
  2. 2Remember a preferenceWrite one durable fact with am memory ingest.
  3. 3Use it laterRetrieve it and see the trace in the console.

1. Install and initialize

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

Plain am init defaults to Hosted Cloud: browser sign-in, Cloud project selection, and a per-installation API key saved in your local profile. No Docker or OpenAI key is required.

If the installer updated PATH, open a new terminal or run:

source "$HOME/.atomicmemory/env"

Already have a Cloud project? Run am init --project <id-or-slug> instead.

2. Remember a travel preference

am memory ingest "I prefer aisle seats when I fly."

The CLI uses the stored project API key — you do not need to export ATOMICMEMORY_API_KEY after init-managed setup.

3. Retrieve it for the next task

am memory search "Which seat should I book?"

Success looks like a result about the aisle-seat preference. That result is the context your application passes to a model before it answers the person.

4. Inspect the result

Open the console at memory.atomicstrata.ai. Under Memories you see the stored content; under Traces you see the ingest and the search, each with its call type, timing, and reconciliation decision.

Something not working? See Troubleshooting.

From an application (SDK)

To wire memory into a Node.js app instead of the CLI, install the client and pass an API key explicitly:

npm install @atomicmemory/sdk
export ATOMICMEMORY_API_URL=https://api.atomicstrata.ai
export ATOMICMEMORY_API_KEY=amc_dev_xxxxxxxxxxxxxxxx

Copy the key from the console under API Keys (shown once at creation), or use the key am init stored locally without exporting it in the CLI path above.

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

const memory = new MemoryClient({
  providers: {
    atomicmemory: {
      apiUrl: process.env.ATOMICMEMORY_API_URL,
      apiKey: process.env.ATOMICMEMORY_API_KEY,
    },
  },
});

await memory.ingest({
  messages: [{ role: 'user', content: 'I prefer aisle seats when I fly.' }],
  scope: { user: 'user_001' },
});

const results = await memory.search({
  query: 'Which seat should I book?',
  scope: { user: 'user_001' },
});

The scope says who this memory belongs to. Every write and every read names a scope, so one user's context never leaks into another's.

What you just built

You completed the smallest durable-memory loop: write a preference, retrieve it for a later request, and inspect the operations that made it happen. Next, see how scopes, corrections, and version history keep that loop safe in a real application: How memory works.

Next steps