Mastra integration

@agentmem/mastra is a drop-in memory layer for Mastra agents. It ships an Integration subclass and two pre-built tools — memorize and remember — that an agent can call without you having to wrap anything.

Install


npm install @agentmem/mastra @mastra/core zod

@mastra/core and zod are peer dependencies (Mastra projects already have them).

Wire it up


import { Agent } from '@mastra/core/agent'
import {
  AgentMemIntegration,
  agentmemMemorize,
  agentmemRemember,
} from '@agentmem/mastra'
 
const agentmem = new AgentMemIntegration({
  apiKey:  process.env.AGENTMEM_API_KEY!,
  agentId: 'support-bot',          // default for every memory operation
  scope:   'team',                 // optional default
})
 
const agent = new Agent({
  name: 'support',
  instructions:
    'You help customers. When you learn a durable fact (preference, decision, ' +
    'constraint), call `memorize`. Before answering, call `remember` to recall ' +
    'prior context.',
  tools: {
    memorize: agentmemMemorize(agentmem),
    remember: agentmemRemember(agentmem),
  },
})

The agent will pick up the tools automatically.

What each tool does

`memorize`

Input	Type	Notes
`statement`	string	Required. The fact to remember.
`agentId`	string	Optional. Override the integration default.
`workflowId`	string	Optional.
`scope`	`private \| team \| global`	Optional.
`role`	`planner \| executor \| observer`	Optional.

Returns a short confirmation string the model can use as a tool result.

`remember`

Input	Type	Notes
`question`	string	Required. Natural-language query.
`agentId`	string	Optional.
`workflowId`	string	Optional.
`topK`	int (1–50)	Default 5.
`scope`	`private \| team \| global`	Optional.

Returns a prompt-prefixed string the agent can splice directly into its reasoning. Empty string if no memories match.

Without the tools

The integration class also exposes direct methods for custom flows:


const text = await agentmem.searchMemory('escalation policy?', { agentId: 'bot' })
const hits = await agentmem.searchMemoryRaw('escalation policy?', { agentId: 'bot' })
await agentmem.createMemory('Customer prefers email.', { agentId: 'bot' })

Defaults and overrides

Merge order: tool input → factory defaults → integration config → error if agentId is still missing.