> ## Documentation Index
> Fetch the complete documentation index at: https://docs.recallium.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Search and recall

> Find prior decisions, fixes, and work history by intent, project, or file

Recallium understands natural-language questions and can narrow results with project and file context.

## Ask the real question

Search for the decision or outcome you need:

> How did we handle expired OAuth tokens, and why did we reject silent refresh?

This is more useful than searching for a single word such as “auth.”

## Search by file

When you know where you are working, include the path:

> What prior decisions and bugs are connected to `src/auth/session.ts`?

> What should I know before changing files under `src/auth/`?

File context can surface relevant history even when you do not remember the title of the earlier decision.

## Search within a project

Name the project when your account contains related repositories:

> What API compatibility decisions did we make in `mobile-app`?

For a broad comparison, ask across the projects you can access:

> Compare the authentication approaches we have used across our projects.

## Narrow an overwhelming result

Add one or more constraints:

* The project
* A directory or file
* The kind of context, such as decisions or bug fixes
* A recent time window
* The workstream or feature name

For example:

> Find recent authentication decisions for files under `src/auth/` in `payment-service`.

## Review the source context

Ask your agent to show which memories support its answer:

> Show the source memories behind that recommendation and open the most relevant one.

This lets you review the original reasoning, connected files, and later decisions that may have replaced it.

## Resume one effort

When you know the workstream, ask for its complete journey:

> Summarize the constraints, decisions, fixes, and latest progress in `checkout-reliability`.

This is often clearer than a broad project search because it follows one body of work.

<Tip>
  If a search result is wrong, add the project and topic before making the query more technically specific.
</Tip>