Core Concepts

The key parts of the Velatir platform and how they fit together.

The Building Blocks

Velatir is built around a small set of concepts. Once these click, the rest of the platform follows.

Organisation & workspaces

Your organisation is the top-level account. Workspaces sit inside it and separate activity by team, department, or project.

Sessions & traces

Every AI interaction is captured as a trace. Related traces are grouped into a session that represents a complete conversation or workflow.

Agents

Gatekeeper and Data Protector review every trace. Coach guides people as they use AI.

Assessments

Each security agent records an assessment for every trace it reviews, with the verdict and the reasoning behind it.

Instructions

Your own rules that tell an agent what to do in a specific scenario, such as always blocking a project codename.

Insights

Aggregated views of how your organisation uses AI, where activity concentrates, and how AI Literacy is tracking.

How They Fit Together

1. A trace is captured

   An AI interaction, in the browser or a desktop app, is captured and sent to Velatir as a trace. Each trace has a direction: Inlet (a request to an AI service), Response (a reply from it), or Signal (an event). Related traces are grouped into a session.

2. Agents assess it

   Gatekeeper and Data Protector review the trace at the same time. Each one records an assessment and reaches a verdict, applying your categories and instructions. Coach is not part of this verdict.

3. The outcome is resolved

   The verdict depends on each security agent’s role. An Observer flags its findings for review. An Enforcer can block the trace. The most restrictive outcome wins.

4. People are notified

   High-criticality findings are sent to your connected channels so the right people know. Access requests and false-positive challenges are handled on the agent, not as a hold on the prompt.

5. Everything is logged

   The trace, each assessment, and the outcome are recorded for audit purposes.

Key Principles

Traces are the foundation

Every AI interaction is captured as a trace. This gives your organisation full visibility into AI usage.

Agents handle the volume

Security agents review every trace automatically. People only see what an agent flags, blocks, or notifies them about. Coach speaks up only at the moments you configure.

Roles control authority

Gatekeeper and Data Protector each run as an Observer or an Enforcer. You can start with monitoring only and add enforcement as you build confidence. Coach has no Observer or Enforcer role.

Instructions tune behaviour

Instructions let you teach an agent how to treat a specific scenario, so its verdicts match your organisation over time.

Everything is auditable

Every trace, assessment, decision, and outcome is logged with full context. Your compliance trail is always complete.

Organisation and Workspaces

Concept	Description
Organisation	Your company account. Holds members, settings, agent defaults, and your subscription.
Workspace	A space within your organisation for a team, department, or project. Workspaces can nest, and each has its own activity.
Organisation roles	Administrator (full access) and Reader (view only).
Workspace roles	Admin, Editor, and Reader, with progressively fewer permissions.

See Roles & permissions for the full breakdown.

Trace Directions

Direction	Description	Example
Inlet	A request going to an AI service	Someone submits a prompt to a chat assistant
Response	A reply coming back from an AI service	The assistant’s answer
Signal	A related event	A session starting, or a background event

Assessment Outcomes

Outcome	What it means
Allowed	The agent had no concern. The trace proceeds.
Flagged	An Observer noted a finding for review, without blocking.
Blocked	An Enforcer stopped the trace.
Escalated	A high-criticality finding also notified your channels. The prompt is not held for a dashboard decision.

---

Understanding agents

Meet Gatekeeper, Data Protector, and Coach.

Trace lifecycle

Follow a trace from capture to final state.