Agentic AI · Company OS

Agents that
run the work.

QUAN 2.0 is an advanced, autonomous agentic operating system: the AI infrastructure that runs your company. Specialized agents reason, plan, and execute real work with persistent memory, real tools, and approval-gated actions, across chat, desktop, and your inbox. Not a chatbot. An operator.

Request access → See how it works
  • Persistent memory
  • Multi-agent teams
  • Human-approved actions
Capabilities

Everything an operator needs,
in one system.

QUAN pairs reasoning with real-world reach. It remembers, plans, uses tools, and asks before anything consequential.

◈

Persistent memory & learning

Typed, editable long-term memory of people, projects, and decisions. It reconciles contradictions and turns outcomes into reusable know-how.

⬡

Multi-agent teams

Specialized agents for sales, support, research, and operations that plan, delegate, and synthesize. A manager and critic keep them honest.

⊕

Real tools & actions

Send email, search the live web, manage a calendar, draft invoices, run code, automate a browser. Dozens of real actions, not just text.

⧉

MCP connectors

Plug in Notion, GitHub, and any Model Context Protocol server. Adding a tool is configuration, not code.

◷

Autonomous workflows

One instruction triggers multi-step execution: plan, act, verify, and self-correct, with a record of every step.

⬢

Omnichannel

The same agents on the web, a native Mac app, Telegram, iMessage, and WhatsApp. Pick up the conversation anywhere.

▣

Sandboxed code execution

Writes and runs Python and JavaScript in an isolated sandbox to compute, transform data, and automate whole pipelines.

⌖

Computer use

Drives a real browser to navigate, fill forms, and operate the web apps that have no API at all.

⊘

Security & offensive tooling

Reconnaissance, vulnerability scanning, and purple-team testing, scope-gated and approval-bound.

✦

Self-extension

Writes, installs, and ships its own tools, skills, and connectors at runtime, no release required.

⟲

Reasoning core

Plans, executes, and verifies its hardest work with a critic before answering; background jobs loop until they converge.

▤

Document intelligence

Reads and extracts from PDF, DOCX, XLSX, images, audio, and video, then acts on what it finds.

◆

Designs & builds UI

Writes real frontend code in its sandbox and deploys it live to a URL, base-path aware, with a shared design system (DESIGN.md). Production frontend, not mockups.

⌗

Knowledge & retrieval

Ingests your documents into a private, searchable knowledge base and recalls the right passage on demand.

◐

Voice & transcription

Reads its answers aloud and turns your speech and audio files into text it can act on.

The full stack

Every skill, in one operator.

QUAN 2.0 ships dozens of real capabilities across cognition, action, and governance. This is what it does in the infrastructure today.

Cognition & memory

  • Persistent typed memory you view, edit, and delete
  • Episodic, procedural, and semantic recall with vectors
  • Reflection, skill promotion, contradiction reconciliation
  • World model, goals, plan / execute / verify reasoning
  • Long-context management and summarization

Multi-agent orchestration

  • Specialized agents: general, sales, CMS, research
  • Manager, planner, and critic roles
  • Parallel delegation, synthesis, adversarial verification
  • Shared blackboard; member-owned agents with scoped tools

Research & the live web

  • Multi-engine web search and deep research synthesis
  • Real-time news, markets, and weather
  • Fetch and parse any URL or HTTP endpoint
  • Knowledge base ingestion and retrieval

Real-world actions

  • Email: read, draft, categorize, send
  • Google Calendar: schedule, reschedule, find openings
  • Invoicing and CRM
  • Image generation; Telegram and social posting
  • Document intelligence: PDF, DOCX, XLSX, CSV, images, audio, video

Code & automation

  • Sandboxed Python and JavaScript execution
  • Browser automation and computer use
  • Live package install into the runtime
  • A private SQL database the agents own and query

Build, deploy & self-extend

  • Writes its own tools, skills, and plugins
  • Designs on-brand UI with a built-in design system (DESIGN.md + impeccable)
  • Hosts what it builds: ships sites and apps to a live URL
  • Scheduled and recurring autonomous jobs
  • Self-healing job resilience

Connectors (MCP)

  • Notion and any Model Context Protocol server
  • New tools added by configuration, not code
  • Connector write actions stay approval-gated

Offensive & defensive security

  • Reconnaissance and OSINT footprinting
  • Network and port scanning, service and version detection
  • TLS / SSL hardening tests and web-app vulnerability scanning
  • On-demand install of pentest tooling, run in an isolated sandbox
  • A full purple-team doctrine: offense informs defense
  • Scope-allowlisted and approval-gated; never aimed at its own infra

Operating constitution (.md)

  • SOUL.md identity and character; AGENT.md operating doctrine
  • USER.md the principal; QUAN.md the house manual
  • SECURITY.md and PENTEST.md trust model and security doctrine
  • memory/people, products, deals, compliance: live business knowledge
  • Hot-reloaded and enforced on every channel; editable in-app
High-grade infrastructure

An engine built for serious work.

Capability is the easy part. QUAN 2.0 is engineered so that power stays reliable, governable, and fast.

Models

Multi-model engine

Routes across OpenAI, Anthropic, Gemini, and local models, reaching for the strongest model for each task and never locked to one vendor.

Reasoning

Plans, then verifies

On hard problems it drafts a plan, executes it, then checks its own work and corrects before answering. Background jobs loop until they converge.

Orchestration

A team with a critic

Agents delegate in parallel and synthesize the results, while an adversarial critic reviews the work. A strict budget keeps fan-out from running away.

Governance

A living constitution

Identity, operating doctrine, and business knowledge live in versioned documents, hot-reloaded and enforced on every channel, editable in seconds.

Runtime

Contained by design

Code and browser automation run sandboxed behind an egress firewall; secrets sit in an AES-256 vault the model can name but never read.

Extensibility

Grows itself

QUAN writes and installs its own tools, skills, and connectors at runtime, and speaks MCP, so a new integration is configuration, not a release.

Model connection routes

Bring your own models.
Route them however you want.

QUAN is model-agnostic. It routes across frontier APIs, your own Claude subscription, open and self-hosted models, and dedicated speech, vision, and image engines, reaching for the right model for each task. These connection routes are configurable, and we add new providers whenever you need them.

Subscription

Claude Pro & Max

Route your existing Claude Pro or Max subscription straight into QUAN, with no per-token API bill.

Frontier APIs

Anthropic & OpenAI

The Claude (Anthropic) API and the OpenAI API as first-class routes, for the strongest reasoning on demand.

Open & emerging

DeepSeek, Moonshot, Kimi

DeepSeek, Moonshot, Kimi, and other emerging providers, wired in as connection routes whenever you want them.

Local & self-hosted

Ollama & your own infra

Ollama and any OpenAI-compatible endpoint: run open models locally or on your own servers, fully private.

Multimodal

Speech, voice & vision

Dedicated speech-to-text, voice, and vision models, so QUAN can hear, read, and see, not only write.

Generative

Image models

Image-generation models for on-brand visuals, routable alongside the language models in the same engine.

Claude Pro / Max Anthropic API OpenAI Ollama DeepSeek Moonshot Kimi Local & self-hosted Speech & voice Vision Image and more, on request
Included

An all-inclusive model catalog, built in. Every plan ships with a curated catalog of all the major models, open-source and frontier alike, including Claude Opus, Sonnet, and Haiku, ready to route under a single monthly subscription.

Free

Or bring your own, and it is free. Run your own self-hosted models, or plug in your own API keys, and QUAN is fully free to use, with no subscription.

How it works

From a goal to finished work.

Give QUAN a goal. This is the loop it runs every time, from understanding to delivery, pausing only where you want a say.

1 · Understand

Reads the goal, recalls the context

It interprets what you asked, then pulls everything relevant from its persistent memory: people, projects, preferences, and past decisions.

2 · Plan

Breaks it into a plan

On anything non-trivial it drafts a plan, decomposing the goal into ordered steps before touching a single tool. Turn on plan mode to review the plan first.

3 · Route

Picks the right model

It chooses the model for the task. Members route through Low, Medium, or Max and QUAN auto-selects the model; admins route across every provider, and can use several models at once.

4 · Act

Uses real tools

It reaches for real tools and connectors: email, calendars, the live web, code in a sandbox, a real browser, and your MCP integrations.

5 · Delegate

Calls in a team

For bigger jobs it fans out to specialized agents in parallel and synthesizes their work, with a manager keeping the effort on budget.

6 · Verify

Checks its own work

An adversarial critic reviews the result, catches errors and gaps, and sends it back for a bounded correction. Background jobs loop until they converge.

7 · Confirm

Waits for your approval

Anything consequential (money, mail, customer records, the outside world) is staged for one tap. Nothing leaves the building without you.

8 · Remember

Learns for next time

It writes the outcome back to memory you own and can edit, and promotes what worked into reusable know-how for the next task.

Everywhere you work

One brain. Every channel.

The full agent stack runs server-side; each surface is just a way in.

▢

Web

A fast, full-featured workspace in the browser.

⌘

Mac app

A native macOS desktop app for your whole team.

➤

Telegram

Message your agents and approve actions on the go.

◌

iMessage

Text QUAN like a colleague, right from Messages.

✆

WhatsApp

Text QUAN on WhatsApp — hand it a task and approve actions by replying.

Built for trust

Powerful by default.
Contained by design.

Capability without the recklessness. The model is treated as untrusted; the guardrails are the product.

  • ⊟
    Hard infrastructure boundaryThe agent can never touch production databases or servers directly.
  • ✓
    Human-in-the-loopEmail, payments, and customer writes are staged for explicit approval.
  • ▣
    Sandboxed executionCode runs isolated, behind an egress firewall, never near your data.
  • ⊠
    Encrypted vaultSecrets are AES-256 encrypted; the model sees names, never values.
  • ▦
    Per-member scopesEvery teammate sees only their own agents and granted tools.
  • ≣
    Full audit trailEvery action the agents take is recorded and reviewable.

Put your operations on autopilot.

Give QUAN 2.0 a goal. Get back finished work, with you in control of every consequential step.

Request access → Explore capabilities