---
url: /relyai-acellemail/features/observability.md
description: >-
  See and audit every AI action — a usage dashboard, full conversation replay,
  tool-call forensics with CSV export, feedback and sentiment, a self-improve
  view, and a read-only permissions matrix.
---

# Observability

RelyAI is built so you never have to take AI on trust. Every request, conversation and tool call is
logged, and a suite of admin surfaces lets you see exactly what the assistant did, how well it
performed, and what it cost. Each surface is its own entry in the **RelyAI** sidebar group.

::: info Logging never stops
Each observability surface can be hidden by turning its feature flag off on
**RelyAI → Settings → Feature Flags** — but hiding a surface only hides the *view*. The underlying
logging keeps running, so nothing is lost while a dashboard is switched off.
:::

## Usage Dashboard

Open **RelyAI → Usage Dashboard** for the health-and-cost picture at a glance:

* **Runs** and **success rate** across your AI activity
* **P95 latency** and **average duration** so you can spot slow tasks
* **Estimated cost** for the period
* A **breakdown by task** to see where the spend and volume go
* **Daily runs** over the last 14 days

![The RelyAI usage dashboard](/relyai-acellemail/img/relyai-observability-usage.png)
*RelyAI → Usage Dashboard*

## Conversations

**RelyAI → Conversations** lists every multi-turn session. Drill into one to **replay the messages,
tool calls and feedback** exactly as they happened — useful for understanding a customer complaint
or confirming what the agent actually did.

## Audit Viewer

**RelyAI → Audit Viewer** is your tool-call forensics record: every agent action, with **filters**,
a **status time-series**, **top failures**, P95 latency and a dedicated **destructive-ops filter**
so you can review just the actions that changed something. Export any filtered view to **CSV** for
your own records or compliance review.

## Feedback

**RelyAI → Feedback** collects **up/down ratings** from across every AI surface, alongside
**sentiment** categorisation, so you can tell at a glance where the assistant is helping and where
it's frustrating people.

## Self-Improve

**RelyAI → Self-Improve** turns all that logging into a to-do list. It clusters **tool failures by
severity**, lists **low-rated conversations** and any **URL hallucinations**, and highlights
**knowledge-base hot zones** — the topics people ask about most where answers fall short — so you
know exactly what to fix next.

## Permissions matrix

On **RelyAI → Settings → Permissions** you'll find a **read-only matrix** of who can use which AI
tool. It's introspection only — it reports what AcelleMail already enforces and never changes it —
and a customer can open a "View my permissions" view of their own access.

## Related pages

* [Agent tools](/relyai-acellemail/features/agent-tools) — the actions the audit viewer records
* [Feature flags and permissions](/relyai-acellemail/guides/feature-flags-and-permissions) — hide a surface without stopping logging