Archive manual

User guide

How to save, query and ingest information in your second brain. Everything on this page is verified against the real server.

Connect your Claude

You do this once. After that your memory is available in any conversation — web, desktop and mobile.

Open claude.ai (or Claude Desktop) and go to Settings → Connectors.
Choose Add custom connector.
Paste this address:
```
https://mybrain.rlz.cl/mcp
```
A window opens for you to sign in: with your email and password, or with Continue with Google.
You will see an authorization screen stating what you are granting (read and write your whole memory). Press Authorize. You are only asked the first time for each connector.
Done. Check it by asking Claude: “what do you have stored about me?”

Using Claude Code in the terminal?

Install the plugin and it sets everything up at once: the connector configured and the /absorber, /guardar and /consultar commands available from any folder.

/plugin marketplace add jpreyestCL/secondbrain
/plugin install secondbrain

The plugin lets you talk to your memory and store single documents. To ingest whole folders you also need the brain command, which is what reads the files on your disk (the server cannot: that is why nothing is uploaded). Two more lines, once:

curl -fsSL https://mybrain.rlz.cl/install.sh | sh
brain login https://mybrain.rlz.cl

The login opens the browser once and links this machine to your account. It asks for no API key: extraction happens here or on the server, never with your own keys in between.

That is it. You can now ask Claude: “absorb the documents in ~/my-deeds” and it prepares the files and stores the facts from each document. The command details are in Bulk ingestion.

If something goes wrong

What you see	What it means and what to do
`invalid_client`	The connector is holding a client id that no longer exists on the server (for example, after access was cleared). Delete the connector in claude.ai and add it again: it registers itself anew.
`account_not_linked`	You tried to sign in with Google using an email that already had a password account, and that email is not verified yet. Check your inbox and confirm it; if it never arrived, resend it from your account.
It sends you back to the login over and over	The session expired (they last 2 days). Sign in again; the connector is still registered.
`403` when using a tool	Your account exists but has no memory assigned. Write to the administrator: your space needs to be provisioned.

To revoke a connector's access at any time, go to your account and press Revoke next to the application.

Talk to your brain

This is the everyday route and needs no terminal: all it takes is having the MCP connector added in Claude.

Save a fact

Just say it in plain language. Claude works out the scope and calls the tool for you.

Remember that my Banco X account is Y

State the relationships explicitly: “the account at Banco X is mine”, “the contract is with such-and-such company”. The graph extracts what you say, not what is implied.
State the real date of the fact, not today's: “since March 2024”. That date travels in reference_time and is what defines when the fact starts being current.

Ingest a document

Attach the file in the chat and ask for it to be ingested. Claude reads it —even scanned PDFs or images, using vision—, splits it and stores it section by section.

Add this contract to my second brain (attached)

Look up

By default it answers with the current state. If you want the history, ask for it: the previous fact was not deleted, it was closed off with its period of validity.

What is my bank account?

Give me the full history of my bank accounts

Never paste passwords or tokens in the chat. The server redacts them before writing anything anyway, but they are better off never leaving your password manager.

The 9 MCP tools

You do not need to call them by hand: Claude uses them for you. They are here so you know exactly what the connector can do with your memory.

Tool	What it does	Parameters
`add_memory`	Saves a fact or a chunk of a document into your memory.	`name, episode_body, source, source_description, group_id, uuid, reference_time`
`search_memory_facts`	Searches facts: what relates to what, and since when.	`query, group_ids, max_facts, center_node_uuid, only_current`
`search_nodes`	Searches people, companies, accounts, places and documents.	`query, max_nodes, entity_types`
`get_episodes`	Lists what was saved most recently, with its original text.	`max_episodes, group_ids`
`get_entity_edge`	Returns one specific fact by its identifier, with its period of validity.	`uuid`
`delete_entity_edge`	Deletes one specific fact (for example, one that was misread).	`uuid`
`delete_episode`	Deletes a stored document and everything inferred from it.	`uuid`
`clear_graph`	Deletes ALL your memory. There is no undo.	`group_ids`
`get_status`	Checks that the server and the database are responding.	`—`

reference_time is ISO-8601 and is the real date of the fact. The server checks it falls between 1900 and at most one year into the future.
only_current defaults to true: it returns only current facts. Set it to false to include the ones that already changed and see the full history.
clear_graph is destructive and has no undo. Export first from /export if you have any doubt at all.

Isolation: the server forces your user's group_id on every call and checks the ownership of every UUID before touching it. Reading or modifying someone else's graph is not possible, not even by passing someone else's group_id by hand.

Bulk ingestion of a folder (the `brain` CLI)

For when you have hundreds of documents on disk and attaching them one by one in the chat makes no sense. Requires a terminal.

Install

One command. It installs whatever is missing and leaves brain available from any folder.

curl -fsSL https://mybrain.rlz.cl/install.sh | sh

Link it to your account

The browser opens once so you can authenticate.

brain login https://mybrain.rlz.cl

You need no API key. Entity extraction happens on the server with its models; on your machine only file reading, OCR and chunking run, and none of those uses a language model.

The pipeline

Replace <slug> with the name of your space. Sign in and the examples come out already filled in.

The steps run in order and each one records its result in the ledger, so you can stop and pick up where you left off.

brain add ~/Documents/inbox

Commands

Command	What it does
`add <folder>`	The whole process in one command. With --review it stops before sending.
`login <url>`	Links this machine to your server account. Done once.
`status`	Shows which stage each document is in.
`scan <folder>`	Only registers the files in the ledger.
`extract`	Only pulls out the text (OCR included where needed).
`classify --auto`	Only assigns domain, type and the real date. No LLM.
`classify --apply <file>`	Applies a manifest you reviewed by hand.
`chunk`	Only splits into sections of the size the graph digests.
`ingest-graph`	Only sends to the graph (flags --doc-id and --force).
`expire`	Marks documents as expired (--all for all of them).

Global flag: --tenant <slug> chooses which space to work on.

Resumable ledger in ~/.brain/<tenant>/: if something fails, you run the command again and it carries on from where it stopped.
OCR for images and scanned PDFs in the extract step.
Credential redaction: passwords, tokens and card numbers are replaced before anything is written to the graph.
Your originals are untouched: the pipeline only reads from the folder.

Ingesting against this server (no SSH)

This is the recommended way, and the only one available if you do not administer the server. The last step travels over the same MCP connector Claude uses: you authenticate with your account and the server does the extraction with its own models.

brain --tenant <slug> ingest-graph --via mcp --url https://mybrain.rlz.cl/mcp

The first time, the browser opens so you can authorize; the token is stored in ~/.brain/<tenant>/. The earlier steps are local.

No terminal: attach the documents in Claude and ask it to save them. Nothing to install.

Export and revoke access

Your memory is yours and comes out whole whenever you want: a JSON with the episodes (original text), the entities and the facts with their period of validity.

Download my memory (JSON)

In your account you can review the open sessions, close them all at once and revoke any OAuth application you have authorized.