Nova Ai-Studio 2.0 is an AI project delivery platform for enterprises and teams.

Think of it as an enterprise AI Agent that can finish projects — not a chatbot that only answers questions.

You state needs in plain language — “run an industry research pack,” “produce a proposal deck.” Nova runs an advanced task cycle (Goal-Loop):

Listen → Analyze → Lock goals → Produce → Strictly accept

It understands the ask, splits clear goals, writes files into the project folder, and can resume if interrupted. Completion is not “the AI said done” — it is checked against a frozen deliverable list: are the files present, and can they open?

The same loop handles a single task or a full project; one person or a collaborating team.

A critical enterprise capability is smart routing: light steps use lite models; critical steps use flagship models. Official tests show enterprise users can save up to about 70% token compute and cost — without always burning the most expensive model. Figures are scenario-dependent; see methodology.

1.1 Who it is for

• Business / brand leads who need research, plans, decks, and report packs — not chat logs
• Collaborating teams that share one project and one deliverable list
• IT / compliance that must govern multi-user access or keep data inside the company
• Procurement & leadership evaluating a product that can ship, collaborate, govern, and control compute cost

1.2 What it is not

• Not a Q&A-only chatbot
• Not a tool that requires installing a stack before first use
• Not limited to one work type — office, research, creative, GEO, media, development, education
• Not locked to public-cloud compute — can connect to customer model services
• Not Amazon Nova / Nova Act — independent product at novapage.online

The practical question: What can Nova help business teams produce?

Start a task in plain language and receive real files you can preview, download, and keep in the project folder — reports, decks, pages, video, social drafts — not a screen of chat. Goal-Loop exists to make those deliverables complete, correct, and finished.

2.1 Capability domains

Capability Hub is organized by business domains that cover day-to-day knowledge work:

2.2 One sentence → deliverables

2.3 Deliverable form matrix

After “what can it do,” organizations care about: can everyone use it, can teams collaborate, can deliverables be verified, where data lives, and whether token compute cost stays under control.

3.0 Smart routing: spend compute where it matters

Long tasks usually include clarifying needs, retrieving sources, drafting, formatting, and checking the list. Always-on flagship pricing rises fast. Nova’s smart routing auto-selects lite vs flagship by step difficulty and importance, with backup keys and fallbacks. Official tests: versus always-on flagship, enterprise users can save up to about 70% tokens/cost (scenario-dependent — methodology). Usage is auditable per user / tenant.

3.1 Five enterprise landing capabilities

3.2 How people start

• Capability Hub / process templates — click “Try it,” natural language launches the task
• Composer — describe the goal; Agent Harness orchestrates tools in the background
• Showcase — browse completed demos on the marketing site before you try

3.3 Agent Harness

Above the model sits Agent Harness: task understanding, Skills/MCP tool orchestration, long-task resume, and deliverable acceptance. Image generation, video, web fetch, and file writes chain into the task directory. Users state the goal; the platform runs the execution framework.

Chapter 2’s capability map, expanded by scenario — plain-language starts, multi-file project packs, smart routing in the background so long tasks need not burn flagship end-to-end.

4.1 Office & documents

4.2 Research & insight

4.3 Content writing

4.4 Marketing & growth (one domain, not the whole product)

4.5 GEO · AI search optimization

4.6 Creative production

4.7 Team collaboration (Team package)

Externally we communicate 400+ capabilities plus ~35 process templates. Discover (main menu) and the composer Capability panel share one Hub — nine major tabs (marketing flywheel, media, GEO·AI search, finance, office, creation, development, brainstorming, education).

5.1 Capability Hub

• L1 category tabs · L2 steppers · L3 pills where needed
• Favorites star · “Try it” smart prefill
• Process templates for multi-stage projects
• Admin Hub visibility controls which tabs/cards appear

5.2 Process templates

Mature chains such as brand-campaign-full, content-flywheel, research-report, GEO audits, PPT/video packs. Templates inject staged write_file contracts so the deliverable list stays honest.

5.3 Model pool

GPT, Claude, Gemini, Qwen, Seedream/Seedance, and more in one pool. Flagship/lite routing, backup keys (≥4), and degradation chains for web/image/video/audio. Model-routing templates save agent + routing + memory choices (keys excluded).

5.4 Launch Registry & Preflight

Capabilities that need style/aspect choices can open Launch / Preflight Studio (platform feature flags). Designers pick templates with real thumbnails; natural-language design intent can be intercepted in Composer when configured.

Team workspace is the collaboration surface for shared files, permissions, versions, and project memory — planned beside personal Projects and General with a third “Team” sidebar tab, isolated by feature flag so personal habits stay intact.

6.1 What is shared

• Project files and task folders
• Conversation context for the collaboration project
• Project memory (admins edit/delete; members read-only)
• Deliverable list acceptance for the shared goal

6.2 Governance

Role-based access, edit locks, and audit-friendly usage. Collaboration UI stays in the left rail and folder — no forced new chrome for personal projects. Sketch authority lives under artifacts/saas-design/team-workspace/.

6.3 Without Team package

Users still see Project | General. Personal SaaS workspaces and General chat remain the default path. Team capabilities can be gray-released and rolled back without breaking personal projects.

Users buy project results, not a verbal “done.” Nova freezes a Session Deliverable Manifest (SDM) at task start and accepts against real on-disk files.

7.1 Goal-Loop + SDM

1. Listen & analyze the user goal
2. Compile slots (numbered list / “must deliver” / Hub profile)
3. Freeze baseline (count + names); raise goalVersion only on explicit change
4. Produce with Agent Harness; write into artifacts/task-*
5. Validate slots ↔ verified paths; repair incomplete items; circuit-break same-gap storms

7.2 Four-line consistency

Assistant body, sticky Results list, task folder, and HTML export must agree. UI trusts engine turn_acceptance_meta and reconciled SDM — not optimistic green from path guesses alone.

7.3 SuperPreview

40+ formats with professional preview; heavy editors (HF Studio, Bento, GrapesJS, tldraw) lazy-load only in edit mode. Paths stay consistent across body links, Results panel, right rail, SuperPreview, and “open task folder.”

7.4 Visual assets & official media

When official imagery is required, Discover → Prepare → Bind → Validate (VAP). Lingering placeholders fail acceptance under binding audit flags. generate_image must not fake official product shots.

Nova supports SaaS multi-tenant and fully private (Docker / air-gapped) with isomorphic capability — pilot in the cloud, migrate inward without rewriting workflows.

8.1 Deployment forms

8.2 Security layers

• JWT auth · tenant isolation · tool permission allowlists
• Workspace sandbox · usage attribution audit
• Session soft-delete / tombstones · no resurrect of deleted chats
• Optional Redis / PostgreSQL control plane; conversation JSONL on disk

8.3 Compute boundary

“Compute stays under customer policy” — models can be Qwen / DeepSeek / local open-source / private endpoints. Zero public-cloud compute dependency is a supported posture when the customer configures it.

9.1 Why enterprises need an Agent that finishes projects

Chat answers questions. Organizations get stuck on multi-file delivery, collaboration, list acceptance, compute cost, and data residency. Nova is an enterprise collaborative Agent: conversation that ships verifiable project deliverables — not a trail of un-archivable chat.

9.2 SMB & mid-market: pilot fast, ship files

Profile: marketing, ops, research, content, sales-enablement teams with many deliverable types.

• Path: SaaS pilot → pick 1–2 real workflows → accept via Results list + folder → then expand seats/scenarios.
• Priority scenarios: market research, decks, GEO/AI-search visibility, content flywheel, light campaign packs.
• Pain relief: no DIY workflow platform; out-of-box Hub + templates; routing controls early token spend.
• When to private: contract/residency requirements, or migrate the same habits on-prem — SaaS and private are isomorphic.

9.3 Large enterprises & groups: governance and audit

Profile: brand HQ, multi-BU matrix, multi-tenant parallelism; need permissions, usage, upgrades, and data boundaries.

• Path: joint IT + business PoC → choose private vs BU SaaS pilot → turn on usage/permission policy → govern Hub visibility by org unit.
• Priority scenarios: cross-BU launch/campaign packs, standardized research + Word/PPT, media planning packs, reusable process templates.
• Pain relief: tenant/workspace isolation; turn queue; list/certificate acceptance for audit; upgrades never overwrite customer data/env; admins govern skills/models, staff consume capabilities.

9.4 Government & public institutions: residency and explainability

Profile: public services, institutional research/comms, SOEs — emphasize data in boundary, explainable process, archivable deliverables.

• Path: prefer full private Docker or air-gapped → customer-held/intranet models → close projects on archivable files + list acceptance, not chat screenshots.
• Priority scenarios: policy/industry research packs, briefing decks, public education content matrices, internal knowledge projects (not Q&A bots only).
• Pain relief: chats, files, control DB can stay on customer premises; no mandatory Nova public compute; offline upgrade + intranet access are standard; align external claims with audit (see Chapter 8).

9.5 Private-deploy decision checklist

Deploy shapes and security layers: Chapter 8 · Private deploy & security. Architecture deep-dive: Chapter 10 · System architecture.

9.6 Data security & trust (application view)

• Residency: private keeps JSONL, artifacts, control DB under customer DATA_ROOT; SaaS is tenant-isolated.
• Isolation: cross-tenant denied; clear project/workspace boundaries; tool execution tighten-able by policy.
• Audit: usage attributed to user/tenant; sessions and deliverable paths support review.
• Trust UX: auto-continue long tasks + list acceptance reduce “we don’t know where it is / dare not give AI our data.”

Security five-layer detail in Chapter 8; competitive selection in Chapter 11 · Competitive comparison.

9.7 Suggested 90-day landing cadence

Figure: enterprise landing cadence (illustrative)

This chapter is for IT and technical evaluation. Business readers can finish Parts 1–3 first and return as needed.

10.1 Logical topology

Think of Nova as “front desk takes the ask → mid-office executes → archive stores deliverables → acceptance checks the list” — Goal-Loop decomposed onto the system.

10.2 Dual tracks

• Conversation resilience — RecoveryBudget dual tracks (recoverable vs hard-fail), Turn Queue for cross-session concurrency
• Deliverable governance — SDM, certificate/shadow/enforce flags, Ground Truth reconcile, repair circuits

10.3 SaaS control plane

Users, tenants, conversation catalog in PG/SQLite; files in cloud-storage hubs; optional OSS for read acceleration. Marketing site is the default public entry; login leads to the workbench.

10.4 Runtime notes for ops

Browser as sole employee entry; async validate; tenant context mandatory; no sync FS blocking on Bridge hot paths; feature flags synchronized across pack / cloud-env / dev launcher.

Compared products include Nova · OpenAI Codex · OpenClaw · Hermes · WorkBuddy · QoderWork · ChatGPT Enterprise · Dify / FastGPT / Coze (also summarized on /en/compare/). Peer products evolve — verify against their latest docs. Nova is unrelated to Amazon Nova / Nova Act.

11.1 Positioning contrast

11.2 When Codex / ChatGPT may suffice

Codex is strong at Agent orchestration, Sites, plugins, and GitHub integration. If the team is deep in OpenAI and accepts App/CLI, it often covers R&D and knowledge work. Nova complements when you also need: browser zero-install for all staff, Team + deliverable list acceptance, customer full-stack Docker private, SaaS→on-prem isomorphic migration, business templates / Capability Hub, and organization-level smart routing cost control.

11.3 One-line difference

Nova Ai-Studio: deliver whole projects through conversation — files you can accept, not a trail of chat. Build workflows/bots elsewhere if that is the primary job; choose Nova when the job is understand → write deliverables → strict accept → govern cost.

12.1 Eight reasons organizations choose Nova

1. Browser zero-install for business users
2. File-level deliverables with list acceptance (Goal-Loop)
3. Smart routing cost control (up to ~70%, scenario-dependent)
4. 400+ capabilities + process templates out of the box
5. Team collaboration path with shared list
6. SaaS and fully private isomorphic deployment
7. SuperPreview + five-path consistency against fake-complete
8. Enterprise security: tenancy, permissions, sandbox, usage audit

12.2 Quick match

12.3 FAQ (selection)

Q1: We already have ChatGPT / Codex — do we still need Nova?
Keep them for chat/dev Agent depth. Add Nova when you need browser-for-all, list acceptance, Team projects, private full-stack, or smart-routing cost governance.

Q2: Is Nova only for marketing AI?
No. Positioning is enterprise / team Agent that can finish projects across office, research, GEO, media, creation, and more.

Q3: How do we choose Nova vs Codex?
Codex for OpenAI-ecosystem Agent orchestration with App/CLI. Nova for org-wide browser delivery, private deploy, Team + SDM acceptance, and routed token cost.

Q4: How do we control token / compute cost?
Enable smart routing; measure against always-on flagship baseline; audit usage per user/tenant. See methodology.

Q5: How do we evaluate in a pilot?
Run a real deliverable task → confirm Results list + folder + export parity → compare token cost on your mix → decide SaaS vs private by residency.

After this paper, treat Nova as an enterprise AI Agent that can finish projects — Goal-Loop of Listen → Analyze → Lock goals → Produce → Strictly accept for solo tasks, full projects, and team delivery.

On the organization side, five landing capabilities: browser zero-install, smart routing cost control, team collaboration, list-based acceptance, private and SaaS isomorphic options. In 2026, Agent competition has moved from “can it chat?” to “can it finish, verify, and govern projects inside the org — at a compute cost you can afford and justify?”

• Core difference: enterprise, team-ready, browser zero-install, smart routing, private/SaaS dual form, out-of-box ease
• Completion rule: deliverable list + real files present — not verbal “done”
• Cost strategy: smart routing per step; up to ~70% token/cost savings; usage auditable
• Experience: business uses the browser; IT can govern and deploy; leaders get auditable project assets
• Landing path: SaaS lowers pilot friction; private matches cloud shape for inward migration

Nova Ai-Studio: the enterprise AI Agent that finishes projects — lower adoption friction, verifiable team delivery, smart routing for compute cost, and controllable data/compute boundaries.

Appendix: login Hero · security narrative · deploy story · Nova Ai-Studio PR · full-case review edition · 2026-07-30

Reading path: know → capabilities → scenarios → reliability → (optional) architecture → selection · 2026-07-30

Related: FAQ · Compare · Token methodology · Showcase · Contact