Key Concepts

Before diving in, let's cover the key concepts that make Cesteral Intelligence work.

Workspaces

A workspace is your working environment. There are two types:

• Personal workspace: Your private space at /workspace
• Organization workspace: Shared space at {org}.yourdomain.com/workspace

All your conversations, governance decisions, knowledge, and automations live within workspaces.

The Workspace Sidebar

The workspace sidebar is one list of pages in four groups. There's no mode toggle and no separate plane to enter or leave: every page is always in the same place.

• Operate — the everyday work: Home, Campaigns (with Pacing nested under it), Automations, and Chat
• Governance — the controls on that work: Queue, Changes, Action results, Policies, and Audit log
• Build — Audiences, Experiments, and Media library
• Workspace — Integrations, Knowledge, and Skills

The Queue row shows a count of pending approvals, Policies shows a dot while a rule is on watch, and Integrations shows one while a connection needs attention or its health couldn't be checked. Incidents and active pauses are reported in the top bar on every page, so you can see when something needs you from wherever you are.

This documentation is organized by lifecycle — Configure → Monitor → Review → Agent — which is how the docs sidebar groups pages. The doc sections don't map one-to-one onto the in-app sidebar groups, but the underlying surfaces are the same: policies and budgets you set under Configure live under Policies in the app's Governance group, the approval queue under Monitor is the Queue in that same group, and so on.

Assistant Conversations

The assistant is the primary interaction model. Rather than navigating menus and forms, you converse with an AI that understands your campaigns, data, and team preferences.

The assistant can:

• Answer questions using your knowledge stack
• Propose campaign actions (launches, budget changes, audience updates)
• Execute approved actions through governed automations
• Learn from your feedback via project memory

Every action the assistant proposes goes through governance evaluation before execution.

Governance

Every proposed action flows through policy evaluation. Key primitives:

• Guardrails — JSONLogic policies that check budget, targeting, brand safety, and more. Managed under Policies > Blocking rules in the app.
• Action manifests — structured diffs showing exactly what will change before execution. Reviewed in the Queue (and bundled changes in Changes).
• Confidence scores — risk-weighted numbers (0–1) that decide what needs human review. Thresholds are configured under Policies > Execution controls.

Actions below your confidence threshold, or that trip a blocking rule, are held in the Queue for manual approval. The full history of every policy decision lives in the Audit log.

Skills

Skills are domain-specific standard operating procedures that encode how your team runs campaigns. Each skill combines orchestration logic with governance rules.

Skills activate in three ways:

You can create custom skills to encode your team's specific workflows. Found under Skills, in the sidebar's Workspace group.

Knowledge

The assistant's context comes from three tiers:

All three are managed under Knowledge, in the sidebar's Workspace group.

Automations

Automations are visual pipelines that process data and trigger governed actions. They consist of nodes connected by edges, with data flowing from entry nodes through processing nodes to outputs. Every automation execution is subject to the same governance controls — guardrails, confidence scoring, and audit logging — as any other platform action.

Found under Automations, in the sidebar's Operate group.

Permissions

Access is controlled through four roles (viewer, editor, admin, owner), each granting a set of resource:action permissions:

See Team Management for the full per-permission breakdown.