Hosted developer preview — now available

Memory infrastructure
you can inspect.

Hosted memory for AI agents — mutations are recorded as traces, retrievals come with inspectable summaries. Scoped, portable, and inspectable end-to-end by design.

Low-latency

retrieval

Traced

mutations

OSS

self-host core

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

const memory = new MemoryClient({
  providers: {
    atomicmemory: {
      apiUrl: 'https://api.atomicstrata.ai',
      apiKey: process.env.ATOMICMEMORY_API_KEY,
    },
  },
})

// Ingest a memory
await memory.ingest({
  messages: [{ role: 'user', content: 'I prefer aisle seats.' }],
  scope: { user: 'user_001' },
})

// Search memory
const results = await memory.search({
  query: 'seat preference',
  scope: { user: 'user_001' },
})

// Capabilities

Memory you can inspect
end-to-end.

Most AI memory tools optimize for recall. AtomicMemory Cloud optimizes for controlled memory state — visible, queryable, and traceable at every layer.

Inspectable memory state

See what the memory layer stores and why. Claims are visible and queryable through the dashboard and API.

Decisions, not blind appends

Core reconciles new facts against existing memory; Cloud records mutation traces — full AUDN decision rendering in progress.

Retrieval traces

Inspect retrieval summaries and returned memory IDs for each query — ranking and exclusion views in progress.

Context packaging

Pack memories into model-ready context with token budgets. See exactly what was included and what was excluded.

Project / user / agent scopes

Isolate memory by scope. Project-scoped isolation with user and agent scoping on API paths.

SDK-portable

Start with the hosted endpoint. Migrate to self-hosted via the open-source Core — same SDK contract; self-host with Postgres/pgvector.

// Deploy your way

Hosted to start. Yours to keep.

Hosted Cloud

Start in minutes. Managed endpoint, zero infrastructure.

  • ✓ Managed API endpoint
  • ✓ Postgres + pgvector provisioned
  • ✓ Managed extraction and embedding pipeline
  • ✓ Usage dashboard included

Self-Hosted

Same SDK contract; self-host the open-source Core with Postgres/pgvector.

  • ✓ Open-source core engine
  • ✓ Docker + Postgres deployment
  • ✓ Full data residency control
  • ✓ No vendor lock-in — migrate any time
View self-host docs

Start with a managed endpoint.
Keep the SDK abstraction.

Ship agents with memory you can actually reason about — then take the core with you when you're ready.