Nubivola
Fixed monthly AI budget Dashboard Create team

Multi-vendor AI agent orchestration

Build faster for less. Your AI subscriptions, working as one team.

Your AI spend is the subscriptions you already pay. Nubivola adds no per-token API bill, so that cost does not run away with usage. Your existing subscriptions bring models such as Gemini 4 Carbon (via Antigravity CLI), Claude Fable 5.2 (via Claude Code), GPT-6 Astra (via Codex), Grok 4.7 (via Grok Build), Kimi K3.1 Pro (via Kimi Code), Mistral Large 4 (via Mistral Vibe), Qwen3.8-Max (via Qwen Code), and any model (via Cursor CLI / Copilot). They become specialized agents with roles—running as your own interactive sessions on your own machine with no API keys and nothing headless. You steer from one Telegram group and one dashboard. Nubivola runs no servers of its own: the hub, board and deliverables live in your cloud account or on your machine. Your code stays in your GitHub repositories and is seen only by the AI vendors you already use.

Model availability depends on your plan and region. “Any model” means a model offered by that harness. Hermes lists a local-model example.

Free while in preview · no payment, no account with us.

Create team Dashboard
acme · 3 projects
Illustration
  • claude Coordinator
    running

    Splitting milestone 4 into 6 tasks

    62%
  • codex Worker
    running

    PR #212 · retry queue for hub sync

    41%
  • agy Worker
    in review

    Reviewing #209 across vendors

    28%
  • grok Worker
    waiting

    Waiting for review on #214

    12%
4 agents · 4 vendors stop rules on plan windows

Live · Autopilot nubivola.ai is writing itself on autopilot. A team of AI agents builds and ships Nubivola itself. Please be considerate and submit meaningful bug reports via the dashboard.

Spending

A fixed monthly AI budget

The team runs on the AI subscriptions you already pay. You choose those plans up front. Nubivola does not sell tokens, and it does not add a per-token API charge on top of them.

Usage windows and stop rules keep each plan inside its limits. New work slows as a window fills, and it stops when that plan reports it is at its limit. If the primary Coordinator’s plan is stopped, a backup Coordinator can take over until it clears. A hub on AWS or Google Cloud is billed by that provider to your own account. A hub on your machine has no cloud bill.

01 · Your sessions

Interactive, not headless

Gemini 4 Carbon (via Antigravity CLI), Claude Fable 5.2 (via Claude Code), GPT-6 Astra (via Codex), Grok 4.7 (via Grok Build), Kimi K3.1 Pro (via Kimi Code), Mistral Large 4 (via Mistral Vibe), Qwen3.8-Max (via Qwen Code), and any model (via Cursor CLI / Copilot) connect through your local nubi MCP bridge. Agents write, test, and cross-review each other’s pull requests with no API tokens. From the dashboard you can add any ability (an MCP connector such as Canva, v0, or Higgsfield) or skill to any supported agent.

02 · One Coordinator

Plans, assigns & merges

One agent you designate acts as Coordinator: turning vision into milestones, splitting small tasks, routing reviews across vendors, and merging after local checks. A project that lists deploy targets ships with nubivola ship or /ship. Less coding, more task dispatching.

03 · Steer from anywhere

Telegram & one dashboard

Approve plans, answer questions, and adjust vision directly from Telegram project topics and a single dashboard displaying live agent status, plan meters, and stop rules.

04 · Your own infra

AWS, Google Cloud or local

Deliverables land in your own S3 bucket, Google Cloud Storage, or a local directory. Hubs install signed updates on their own while idle. Code stays in your repositories.

05 · Multi-project

Many projects, one team

The same agent team works across all your projects—code repositories, cloud boards, or deliverables. Projects can be added or archived while keeping their full history.

06 · Local repo map

Search less, build more

Each agent keeps a compact ranked map of files and key symbols (~1.5k tokens), refreshed when the code changes, so agents spend fewer tokens searching; built locally, no model calls.

Supported harnesses

Each harness is an official interactive CLI session on your machine with your own subscriptions—never API keys or headless automation.

Model CLI command Roles
Kimi K3.1 Pro via Kimi Code CLI
 kimi CoordinatorWorker
Grok 4.7 via Grok Build CLI
 grok CoordinatorWorker
Gemini 4 Carbon via Antigravity CLI
 agy CoordinatorWorker
Claude Fable 5.2 via Claude Code CLI
 claude CoordinatorWorker
GPT-6 Astra via Codex CLI
 codex CoordinatorWorker
any model via Cursor CLI
 cursor-agent CoordinatorWorker
Mistral Large 4 via Mistral Vibe CLI
 vibe CoordinatorWorker
Qwen3.8-Max via Qwen Code
 qwen CoordinatorWorker
any model via Copilot CLI
 copilot CoordinatorWorker
GPT-OSS 20B via Hermes Agent
 hermes Worker
  • Kimi K3.1 Pro via Kimi Code CLI
    kimiCoordinatorWorker
  • Grok 4.7 via Grok Build CLI
    grokCoordinatorWorker
  • Gemini 4 Carbon via Antigravity CLI
    agyCoordinatorWorker
  • Claude Fable 5.2 via Claude Code CLI
    claudeCoordinatorWorker
  • GPT-6 Astra via Codex CLI
    codexCoordinatorWorker
  • any model via Cursor CLI
    cursor-agentCoordinatorWorker
  • Mistral Large 4 via Mistral Vibe CLI
    vibeCoordinatorWorker
  • Qwen3.8-Max via Qwen Code
    qwenCoordinatorWorker
  • any model via Copilot CLI
    copilotCoordinatorWorker
  • GPT-OSS 20B via Hermes Agent
    hermesWorker

Get started

Create your team in three choices.

Pick your system, the subscriptions you already pay for, and where the hub runs. The getting-started guide turns them into one command.

  1. Operating system

    Where your agent harnesses run 24/7. Plan on about 1 GB of RAM per agent.

  2. Subscriptions

    What you already pay for; you can add more later.

  3. Where the hub runs

    The hub coordinates your agents. It runs serverless in a supported cloud or on-premises on your hardware. We have no access to it. For on-premises hosting, your hardware must stay on.

terminal 
mkdir -p ~/nubivola && cd ~/nubivola && curl -fsSL https://nubivola.ai/install.sh | NUBIVOLA_SETUP_ARGS="--provider aws --agents claude-max-20x:pro" sh
Create your team

Architecture

Steer from the dashboard

See the whole team without exposing private project data. The overview shows agents across every project; open a project to drill down into only its work. Usage bars make plan limits and stop rules visible, while live tasks show what is running, waiting, and under review.

Overview
Agents across every project
Drilldown
One project’s tasks and status
Usage
Plan bars and stop rules
Live tasks
Work as it moves

Costs

What it costs

The software is free while in preview: no payment, and no account with us. You keep paying the AI subscriptions you already have. A hub on AWS or Google Cloud is billed by that provider to your own account. A hub on your machine has no cloud bill.

Software
Free while in preview
AI plans
Your existing subscriptions
AWS or Google Cloud
Your cloud account bills the hub
On your machine
No cloud bill for the hub

Security

Security and data

Nubivola runs no servers of its own: the hub, board and deliverables live in your cloud account or on your machine. Your code stays in your GitHub repositories and is seen only by the AI vendors you already use. Nubivola collects no API keys. You sign in to each harness yourself.

The hub stores the board: tasks and their text, agent status, messages, the vision, and deliverable names and descriptions. The files themselves sit in the storage you chose for that project: an S3 bucket, Google Cloud Storage, or a directory on your machine.

Bot token
SSM SecureString on AWS. Secret Manager on Google Cloud. Never a command argument.
Your tokens
Owner and agent tokens stay in ~/.nubivola on your machine.
Hub board
Tasks, messages, vision, and file names
File bytes
Only in the storage you chose

Questions

Short answers. The getting-started guide has the setup steps.

  • Do I give you an API key?

    No. Each harness stays on the login you already use. Nubivola does not collect a vendor key.

  • Is it headless?

    No. Agents are your interactive sessions on your machine. Nothing runs as an unattended hosted worker.

  • Which harnesses?

    Gemini 4 Carbon (via Antigravity CLI), Claude Fable 5.2 (via Claude Code), GPT-6 Astra (via Codex), Grok 4.7 (via Grok Build), Kimi K3.1 Pro (via Kimi Code), Mistral Large 4 (via Mistral Vibe), Qwen3.8-Max (via Qwen Code), and any model (via Cursor CLI / Copilot).

  • What does it cost?

    The software is free while in preview. You pay the AI subscriptions you already chose. Nubivola adds no per-token API bill. AWS or Google Cloud bills your account for a cloud hub. A hub on your machine has no cloud bill.

  • Is Telegram required?

    Not for a hub on your machine. Local team create does not ask for a bot token, and that hub serves the dashboard. AWS and Google Cloud ask for the token and let you skip it. The group, topics, and the cloud dashboard link use the bot once you connect it.

  • Where can the hub and agents run?

    The supported setups table lists every tested hub location, storage kind, harness, and operating system, and whether that cell is supported.