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

# Agent Operations

See what agents did, decide what happens next, keep the evidence — human supervision built into the platform (private beta).

> **See what your agents did, decide what happens next, and keep the evidence.** Agent Operations is the operating view over every governed run — human supervision built into the platform, not bolted on after.

> **Availability: operator-assisted during the invite-only private beta.** The Agent Operations views are not yet switched on for self-service use — our team runs them with you and brings you the runs, exceptions, and evidence they surface. This page describes the supervision model your work is governed by; treat it as forward documentation of the self-service surface, not a view you can open on your own right now. [Request access](https://mediafier.ai/request-access) to start.

Exact controls and features vary by organization and are authoritative in the authenticated product.

***

## What it does

* **Shows what agents and workflows are doing** — runs, batches, MediaClaws, and the tools they called, as they happen and afterwards.
* **Surfaces exceptions and human interventions** — the work that needs a person is brought to the front rather than buried in a log.
* **Connects actions to evidence and outcomes** — every run carries its evidence, its cost, and its result, correlated by the same trace identifier the gateway returns on every call.
* **Helps operators evaluate proposed improvements** — when an agent or a MediaClaw proposes a change to how it works, the proposal is reviewed against evidence before it becomes active.
* **Makes approved learning visible over time** — what changed, when, on whose decision, and what it did to outcomes.

## What it enables

* **Human-in-the-loop supervision.** Operators approve, correct, or decline work where the organization has decided a person must.
* **Investigation of failed or unusual work.** From an exception to the run, the tool calls, the evidence, and the audit record — without leaving the platform.
* **Evaluation before changing agent behaviour.** Nothing an agent learns changes production behaviour until it has been reviewed and validated.
* **Evidence-backed improvement.** Improvements are measured against the organization's own approved baselines, not asserted.
* **Operational intelligence across repeated runs.** Patterns across runs — what fails, what costs, what a person had to touch — become the input to the next improvement.

## Where the authority is

The exact actions, review states, workflow controls, attention categories, and available operations are shown in the authenticated Agent Operations surface, and they are reachable programmatically through the same governed endpoint your agents already use — discover them with `tools/list` once you are signed in. Public documentation describes what Agent Operations is for; the authenticated surface is the contract for what you can do today.

## What it is not

* It is not a log viewer. Logs exist underneath it; Agent Operations is the decision surface built on the evidence.
* It is not autonomous promotion. Proposed improvements wait for a human.
* It is not cross-organization. You see your organization's work; nothing from another tenant appears here (see [Learning & Data Boundaries](/trust/learning-and-data-boundaries.md)).

## Where to go next

| If you're…                                          | Read next                                                                           |
| --------------------------------------------------- | ----------------------------------------------------------------------------------- |
| Asking what an outcome cost and whether it improved | [Spend & Outcome Evidence](/overview/spend-and-outcome-evidence.md)                 |
| Asking what the platform learns from your work      | [Learning & Data Boundaries](/trust/learning-and-data-boundaries.md)                |
| Following a single call end to end                  | [Audit, Observability & Compliance](/trust/audit-observability-and-compliance.md)   |
| Understanding how agents get better                 | [Decision Tracing & Learning Loop](/platform/decision-tracing-and-learning-loop.md) |