Explore memoryLong-term memory for AI agents. Save project context, preferences and decisions, then retrieve what matters in the next session.
Connected agents can use the same saved context.
Try it in this demo: select an agent and save a sample.
Select another to see the same memory.
Sample text stays in this page. Save it, then select another agent.
Use the SDK in your code, connect an MCP client, or choose encrypted memory on Walrus. Start with the setup that fits your application.
npm install @lighthouse-ai/core @lighthouse-ai/engine-batched @lighthouse-ai/store-lighthouse @lighthouse-ai/embed-localimport { createStorage, createEmbedder } from '@lighthouse-ai/core';
import { BatchedEngine } from '@lighthouse-ai/engine-batched';
import '@lighthouse-ai/store-lighthouse';
import '@lighthouse-ai/embed-local';
// Connect your agent to persistent memory.
const memory = new BatchedEngine(
await createStorage('lh-ipfs-filecoin', {
apiKey: process.env.LIGHTHOUSE_API_KEY,
}),
{ namespace: 'my-agent', embedder: await createEmbedder('local') }
);
// Save context and persist it to the network.
await memory.remember('Our project uses TypeScript.', {
tags: ['project'],
});
await memory.flush();
// Retrieve the context when your agent needs it.
const context = await memory.recall('What language do we use?');
console.log(context);
Set LIGHTHOUSE_API_KEY in your environment. This example stores public, unencrypted memory.
Read the guide
Give your agent a persistent brain through the Memory SDK or a supported MCP connection. Save the context you choose and retrieve relevant records in later sessions. Each application needs its own connection.
Connect through MCPClaude
CodexUse decentralised storage for your memory records.
Decide what to save, how to retrieve it and what to share.
Save project facts, preferences and decisions outside the conversation. Your application can search these records and add relevant context to a later request.
Stored memory batches have content identifiers, or CIDs. Use them to check that retrieved content matches the saved version. This checks integrity, not whether a statement is true.
bafy…7k2mbafy…7k2mYour application decides which records to save and pass to each agent. Tags and namespaces organise context; encryption and access controls need separate configuration.
How memory works, what is shared and which setup to choose.
A persistent brain is the memory your agent can return to across sessions. Lighthouse stores and retrieves the context your application chooses to save. Your AI model handles reasoning and responses; Lighthouse provides the memory behind them.
Use the Memory SDK in your application, or connect a supported client through Model Context Protocol (MCP). The setup guide covers the connection and credentials each option needs.
Connected agents can use a shared memory backend. Each application needs a supported integration and the right configuration. Switching models alone does not connect their existing chat histories.
No. Batched memory and index snapshots are unencrypted. Memwal encrypts its Walrus blobs with SEAL, but optional IPFS mirrors are public. Choose the setup before storing sensitive information. See the encryption details.
Content identifiers, or CIDs, let you check that retrieved records match the saved content. They check content integrity. They do not establish whether a statement is true or an agent’s answer is correct.
Not necessarily. Forgetting and deletion depend on the memory engine. Removing a record from an index does not erase an existing stored blob. Plan retention and deletion for the storage option you use.
START WITH LIGHTHOUSE MEMORY