Prefactor Overview

Prefactor is an AI governance platform that helps you understand, manage, and control how your AI agents behave in production.

Prefactor’s web platform gives you a structured view of:

Who is running AI agents (accounts, teams, environments).
What those agents are (individual agents and their versions).
How they behave over time (instances, spans, and schemas).

Once you’ve created or joined an account and selected an environment, the web platform gives you a single place to:

Register and manage your agents.
Inspect how they run over time.
Track changes to their code and behaviour.
Define the structure of the activity they produce.
Control access, environments, and programmatic integration.

Accounts and environments

When you sign in, you land in a Prefactor account. An account represents the organisation, project, or customer whose agents you are governing.

Within an account, you can:

Configure Account settings, including the account name, team, invites, API tokens, and environments.
Use Environments (for example, Development, Staging, Production) to separate activity and configuration for different stages of your lifecycle, while keeping them under the same account.

The environment selector in the header controls which agents and activity you are currently viewing across the platform.

Agents

An agent is any application or service that you have integrated with Prefactor.

From the Agents page you can:

See a 24‑hour view of activity across all agents in the current environment, broken down by state (Pending, Active, Complete, Failed, Cancelled).
Browse a list of agents, along with their current status and descriptions.
Register new agents as you integrate more systems.

Each agent has its own set of pages that let you examine its behaviour in more detail and apply governance controls at the level of a single agent.

Agent overview

The Overview page for an agent gives you a high‑level picture of how that agent is behaving.

You can see:

The agent’s status, description, and identifying information.
When it was registered and last updated.
Activity over the last 24 hours, summarised by span state.
A concise summary of how many instances are Pending, Active, Completed, or failed.

From here you can drill into instances, versions, and the activity schema to understand both what the agent is doing and how it has evolved over time.

Agent instances

Each time an agent runs, Prefactor records an instance: a single execution of a specific version of that agent.

The Instances page for an agent shows:

A table of all instances in the current environment.
The status of each run (Pending, Active, Complete).
The time taken and when the run started and finished.
Which version of the agent was responsible for that run.

From this page you can open a particular instance to see what happened during that execution and how it contributed to the agent’s overall behaviour.

Agent instance views

For a single instance, the platform provides two complementary views:

Activity – lists the spans produced during that run, such as agent‑level spans, language model calls, and tool invocations, along with their duration, status, and timing.
Details – shows metadata about the run: the agent and version, lifecycle timestamps, and current status.

These views help you understand the concrete steps an agent took during a specific execution and how long each step took, which is essential when assessing compliance with internal policies or service expectations.

Agent versions

As you deploy new code, Prefactor tracks each version of an agent.

The Versions page shows:

Version history for the agent, including when each version was first seen.
Which version is currently set as the active one.
Version‑specific details such as the version identifier and associated schema version.
Recent instances for the selected version, so you can see how that release is behaving in practice.

By tying instances back to versions, Prefactor helps you reason about the impact of code changes on agent behaviour and gives you clear anchors for rollout and rollback decisions.

Activity schema

To make agent behaviour understandable and enforceable, Prefactor uses an activity schema to describe the structure of the spans your agents produce.

The Activity schema page for an agent shows:

The available schema versions for that agent and which one is current.
The span types covered by the schema (for example, agent‑level spans, chains, language model calls, and tools).
JSON Schema definitions for each span type’s payload and result.

By defining and versioning these schemas, you can:

Ensure that agent activity conforms to expected structures.
Gradually tighten validation as your integration matures.
Build reliable governance controls on top of consistent, structured data.

Account settings

The Account settings area collects all of the top‑level controls for an account:

Details – the account’s name and unique identifier.
Team – the list of people who have access to the account.
Invites – outstanding invitations for new team members.
API tokens – tokens used for programmatic access, including account‑ and environment‑scoped tokens that you can create, suspend, and revoke.
Environments – the environments defined for the account, along with their IDs and actions to switch, edit, or delete them.

Together, these pages let you manage who can see and change your AI agents, how they authenticate to Prefactor, and how their activity is separated between different stages of your lifecycle.