# See what ReacherX △ Agent is doing

Use Agent observability to inspect searches, memories, quality checks, and activity when results need a closer look.

By Salman · 2026-07-07 · tutorials

Canonical: https://reacherx.com/blog/understand-agent-observability

“Observability” is a technical word for being able to see what a system is doing. In ReacherX, the Agent observability page lets you look behind the results.

You don't need to be a developer to use it. If you want to know what △ Agent searched for, what it remembered, or why some work needs attention, this is where to look.

## Start with a question

Suppose △ Agent keeps finding people who are too far from your target audience. Open the Discovery tab and inspect the search queries. Do those searches describe the people you wanted?

You can search and filter the inventory, then open an individual record for details. Query status matters: not every query △ Agent writes ends up being run. Duplicate or low-novelty proposals may be rejected.

If the search looks right but the people don't, read their qualification explanations too. The problem may be how fit was judged, rather than how people were found.

## Check what △ Agent remembers

The Memory tab lets you inspect saved knowledge and instructions. Look at the source, status, and supporting information. A direct instruction from you carries different meaning from a lesson inferred by the system.

A confidence number is the system's assessment, not a guarantee. Ask whether the memory is supported and still relevant to the current workspace.

If you recently changed your audience or product, look for instructions that no longer apply. Tell △ Agent what should replace them, then check the saved result. [This guide](/blog/teach-reacherx-what-you-want) gives examples of useful corrections.

Interactive demo: Inspect discovery and memory. Open a search query to see what △ Agent found.

## Follow the activity behind a result

The page includes Overview, Discovery, Quality, Memory, and Activity tabs. Activity records can lead to details about an event, an evaluator run (an automatic quality check), or a memory suggestion.

Those distinctions help when you're investigating learning. An event happening, a lesson being suggested, and a memory being promoted (a suggestion turned into a saved instruction) are separate steps. A suggested lesson hasn't necessarily become an instruction △ Agent will use.

You can filter the view and use the date range to narrow the investigation. If something failed, record the visible reason and what you were doing before retrying. That makes a useful bug report.

## Use the two dashboards together

Start with [Analytics](/blog/read-your-reacherx-analytics) when you want to review outcomes and work waiting for you. Use Agent observability when you want to inspect how that work happened.

For example: Analytics shows few qualified people. Discovery shows broad searches. A person's qualification explanation shows missing evidence. Together, those details give you a more useful question to ask than “Why is AI bad today?”

If you need help interpreting a result, [join Discord](https://discord.gg/BQttyr8jY) and describe the workspace's purpose and the record you inspected. Share a redacted screenshot. Developers can continue with [how the reporting system works](/blog/how-reacherx-reporting-works).