> For the complete documentation index, see [llms.txt](https://docs.cleric.ai/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.cleric.ai/learning/how-cleric-learns.md).

# How Cleric Learns

How Cleric improves over time

Cleric continuously improves its understanding of your environment through several mechanisms.

## Calibration

Cleric begins calibrating to your environment as soon as you connect your first integration. It analyzes the data sources to distill working knowledge it can draw on during investigations. Calibration tasks fall into two groups:

* **Onboarding tasks**: Run during initial setup with the Cleric team and do not refresh on a schedule.
* **Discovery tasks**: Run against your live infrastructure to build the service catalog and observability conventions. The service catalog discovery task re-runs automatically once a day so the catalog reflects your current infrastructure.

You can review what Cleric has analyzed on the **System Health** page in the sidebar.

## Passive Context Collection

Cleric automatically gathers context from:

* **Alert structure**: Field names, severity indicators, alert groups
* **Linked resources**: Dashboard URLs, runbook links in alert definitions
* **Naming patterns**: Service naming conventions, metric patterns
* **Infrastructure relationships**: Service dependencies, pod ownership

The more issues Cleric investigates, the better it understands these patterns.

## Memories

When you share facts about your environment during investigations, Cleric offers to remember them. [Memories](/learning/memories.md) capture factual knowledge like service dependencies, environment configurations, and known alert patterns.

Memories are automatically recalled during future investigations when relevant.

## Topics

Cleric maintains a written overview of your environment, organized into areas like alerts, architecture, observability, and deploys. Each Topic is a summary you can read to see what Cleric knows about that part of your system, and every claim links to the source it came from so you can verify it.

This gives your team a standing reference instead of re-deriving how things work under pressure: which alerts are noise, where a service's logs live, how a release reaches production. Cleric writes each summary from the facts and memories it has gathered and checks them against live data sources, so the overview reflects your actual environment rather than out-of-date documentation.

Because Cleric refreshes its topic summaries in the background, a topic can lag recent changes, and one that Cleric has gathered little about may be sparse or empty.

Browse topics under **Knowledge > Topics** in the sidebar.

## Feedback

Your interactions with Cleric provide valuable learning signals. Cleric learns from three types of feedback:

* **Explicit Feedback:** Direct input you provide about Cleric's performance, including message ratings and corrections you share during conversations.
* **Implicit Feedback:** Actions taken during and after investigations. When you follow Cleric's recommendations, ask follow-up questions, or pivot to a different area, these actions help Cleric understand what approaches are effective.
* **Conversation Analysis:** Cleric analyzes past investigations to identify patterns. How your team discusses results, which areas require clarification, and how issues ultimately get resolved all inform future investigations.

### Message Ratings

Rate any Cleric message on a 1-5 scale. This helps Cleric understand which responses were valuable and which missed the mark.

**Rate liberally, especially low ratings.** When a response misses the mark, that feedback helps identify where investigations go wrong. Patterns in low ratings drive improvements to how Cleric approaches similar situations.

If Cleric missed something or took the wrong approach, tell it directly in the conversation. Cleric learns from corrections and can adjust its investigation in real time.