Introduction

Nova Admin started as a support dashboard for Nova and grew into a full agent orchestration platform. It is the control plane that operators use to compose AI agents, wire them to real tools, schedule them on crons or events, and inspect every run.

The mental model

Agents ── perform tasks using tools and memory
Skills ── reusable prompt fragments any agent can invoke
Flows ──── multi-step workflows on a cron, manual trigger, or event
Sandboxes ── isolated execution environments (browsers, terminals, desktops)
Integrations ── credentialled bridges to external services
Observability ── traces, timelines, and failure debugging for every run

Everything is stored in Firestore. Agents, skills, flows, and integrations are first-class records you can edit live; the runtime picks up the change on the next invocation.

Who this is for

• Operators running agents against real production data and external services.
• Developers extending the platform with new providers, actions, or skills.
• Reviewers who need to audit what an agent did, what tool it called, and what it returned.

What's in these docs

• Getting Started — an end-to-end tour of the dashboard.
• Guides — hand-written developer guides: Authentication, App artifacts & prototypes, and Access & sharing.
• API Reference — every REST endpoint, generated from the OpenAPI spec.
• Concepts — the platform primitives (agents, flows, skills, sandboxes, integrations, observability).
• Integrations — per-provider setup guides plus the generated Integration catalog.

The API Reference and Integration catalog are generated from the codebase (bun run docs:generate), so they reflect how the platform actually behaves today. There are no aspirational sections. Anything not yet shipped is explicitly marked Preview.