Start using Brain

Connect a source to Brain, run its first ingestion, and search what your product now remembers. Brain is available in every Lepsto product; there is nothing to install.

This tutorial uses the Lepsto web app. Agents can also work with Brain over MCP or REST.

Before you start

You need access to a Lepsto product and material in GitHub, Google Drive, or ClickUp. For a private source, an organization owner or admin must install a connection and attach it to your product. A public GitHub repository needs no connection.

Ingestion uses language-model and embedding tokens. The first run reads everything in scope, so start with material you want this product to remember.

1. Open Brain

Open your product in the web app and select Brain in its navigation. You arrive at the search page. A product with no memories has no results yet.

Brain keeps each product’s memory separate. Connected sources write to the product’s shared pool; agents can also keep private working notes.

2. Make your source available

On the web app’s Integrations pages, have an organization owner or admin install the source connection and attach it to your product. If you don’t have that role, request the connection.

For a public GitHub repository, skip this step. For other sources, check what each connector reads before proceeding. Connecting grants access; it does not import content.

3. Create a daemon

Open Brain’s Ingest page and create a daemon: a source that Brain reads on a schedule.

Select the source type and its scope:

• GitHub: one repository and one branch.
• Google Drive: a connected account and the files and folders picked for it.
• ClickUp: one workspace, optionally narrowed to a space or list, plus any chat channels you select.

Choose the content categories to ingest. You can narrow GitHub Markdown with include and exclude patterns. Source trust sets the authority of imported memories; load mode controls immediate or deferred processing. See daemon settings for the defaults and trade-offs.

4. Run the first ingestion

A new daemon starts enabled. Select Run now on Ingest, or wait for the hourly cycle.

Open the daemon’s page to inspect the run and its items. A large first ingestion or deferred batch can take hours. Later runs skip unchanged content. A daemon with a run in progress is not started again by the hourly cycle.

A done run can still contain denied items, so check item outcomes as well as run status. Fix source access if an expected item was denied. A failed run has a retryable failure; the next cycle tries again.

5. Verify it works

Open Browse and check that memories from your source appear. The counts at the top give you a second check that material was stored.

Return to search and ask about a statement you know the source contains. For example, if your product’s release notes record a checkout decision, ask, “Why did we change checkout?” Open a result and compare its text with the source.

Search matches meaning, so the question need not repeat the source’s wording. A new product usually has direct Results before it has Profile or Proactive results. Cognition fills those groups as it consolidates shared memories.

Add something through an agent

You can also ask an agent to store a memory. The web app reads, corrects, and deletes memories; it does not create them. A single write stores the original text and can produce several derived facts.

Use Brain from an agent or backend

For example, brain_memory_search searches memory and brain_memory_add stores it. Use a credential scoped to the product and access needed by the operation. See MCP setup, API keys, and the Brain REST reference.

Both programmatic surfaces require a platform credential and are for agents or your backend. Brain has no public browser API for a publishable key, and @lepsto/sdk has no Brain subpath.

If nothing appears