Nova is 24 specialist agents plus an automation layer that reacts to events — webhooks, metrics, connector events — and runs the repeatable work in the background: playbooks, durable processes, a knowledge base you feed. Reach it from Telegram, Slack, Discord, or your terminal; run it on your own model subscriptions and your own machine. It prepares freely, then asks before anything ships, spends, or sends.
Ready to execute: publish the page, schedule 3 emails, launch ads at $400 total. Nothing has gone live yet.
approved · executing… links in thread ✓No magic router, no black box. Whether it starts as a message you send or an event that fires an automation, a request is classified, decomposed if it's big, and executed in two phases — with the consequential half gated behind you.
Short messages skip straight to the model. Repeat requests hit a pattern cache of plans that worked before. Only genuinely new, complex asks pay for LLM classification.
Complex requests become a plan: subtasks, dependency order, and a specialist agent for each. Independent branches run in parallel.
Research, drafts, and images happen freely. Publishing, sending, and spending wait for the approval card — Approve, Revise, or Cancel, from your chat.
Nova connects to your email, calendar, CRM, and ad accounts. The security posture treats that seriously.
Nova is event-driven and process-durable: feed it your knowledge, wire it to your tools, and let it run the repeatable work in the background — every consequential step still passing the same approval gate.
Each agent has its own system prompt, tools, skills, and MCP access — a working team weighted toward marketing and operations, and fully yours to edit. Adding an agent is writing a markdown file.
An optional distributed layer: seven executive nodes, each with its own reasoning persona, deliberating over a shared database. Independent analysis, an adversarial pre-mortem, then synthesized options with confidence scores — you pick, it executes.
/board should we switch to usage-based pricing?
Run on your Claude, Gemini, or Codex subscriptions via their CLIs, or add any OpenAI-compatible model — OpenRouter, OpenAI, local Ollama or vLLM — with one command. Subscription-first, your keys.
One nova command runs everything, and nova connect drops you into your running Nova from any terminal — local or VPS — with live agent activity and inline approvals.
Feed it PDFs, docs, and URLs — chunked and embedded on your machine, retrieved across every agent with source citations. Personal, team, or per-agent.
Facts, goals, and tasks with local vector search — Nova remembers across conversations and injects the right context.
Successful plans are cached and reused; proven wins are promoted into learned skills over time.
Morning briefings, smart check-ins, recurring and conditional tasks — it works while you don't.
MCP servers (Notion, Google Workspace, Playwright, Cloudflare, GoHighLevel) plus two-way business connectors — Stripe, Shopify, Zendesk, HubSpot — on per-user credentials.
Agents discover tools on demand from the shell instead of loading every schema into the prompt, so the toolset grows without bloating context. Nova's own capabilities work the same way.
Query the systems your numbers live in — an HTTP endpoint, a read-only SQLite file, or a connector — for reports and for agents mid-task.
Earned-autonomy ladder, spend caps and approval matrices, role-based permissions, encrypted secrets, and a full action ledger with ROI reporting.
Image generation, DOCX/XLSX/PPTX/PDF creation, research writing, ad extraction — invoked by agents as needed.
Inbound and outbound calls via Twilio; voice messages transcribed with Groq or local Whisper.
One line clones Nova, sets up Bun and the Claude Code CLI for you, then a guided wizard connects Telegram and your AI provider — no file editing, and it detects your Telegram user ID automatically. All you need first is a bot token from @BotFather. Prefer to clone yourself? git clone https://github.com/djbelieny/nova && cd nova && bash bootstrap.sh. Everything else is optional and documented in the repo.
Not everyone has the time — or a spare engineer — to stand up their own AI team. If you want Nova running for your business but don't want to build it yourself, work with me directly: done-for-you setup, consulting & advisory, and custom agents & integrations built around how you actually work. Paid engagements, scoped to what you need.
Booked directly with me, Jake Belieny.