In development — Phase 1

Agency OS — the workspace for creative teams

A native desktop application that combines project management, brand intelligence, team communication, and multi-model AI in one unified experience. Built with Tauri 2, React, and Supabase. macOS first.

Concept

One app instead of twelve

Creative agencies run on Slack, Notion, Figma, Google Drive, Asana, invoicing tools, and scattered spreadsheets. Agency OS replaces this fragmented workflow with a single native application.

Not a PM tool with AI bolted on

Agency OS is a brand intelligence platform with operations and real-time communication attached. The AI knows your brand voice, personas, and guidelines — every output is on-brand by default.

Native desktop, not another browser tab

Tauri 2 delivers a 3–8 MB binary that launches in under 0.5 seconds and uses 30–50 MB RAM. No Electron bloat, no web app lag. System notifications, file system access, offline resilience.

Obsidian meets aWork

Knowledge management with the depth of Obsidian — Markdown docs, linked references, brand knowledge base. Task delegation and team coordination with the structure of aWork — kanban, calendar, time tracking.

Multi-model AI via OpenRouter

Every team member chooses their preferred AI model — Claude, GPT, Gemini, Llama. One API, one billing, all models. Brand context is injected into every prompt regardless of which model responds.

Modules

Seven core modules

Each module is accessible from a circular scroll-to-navigate menu. Sub-items appear as orbital cards around the center ring.

◉

Tasks

Kanban boards with drag-and-drop. Time tracking per task. Delegate to team members or freelancers. Priority levels, due dates, file attachments.

Time tracking Delegate
◎

Chat

Project-linked channels with threaded messages. Direct messages. Voice and video calls via LiveKit. @ai mentions for instant AI assistance in any conversation.

Realtime LiveKit
◈

Plan

Calendar view across projects. Governance and approval workflows. Kanban at the project level — track milestones, not just tasks.

Calendar Governance
◇

Files & Brand

Brand assets vault with version history. Brand core: voice, personas, positioning, guidelines. Competitor intelligence. Image and file storage via Supabase Storage.

Brand core Personas Competitor
◆

Create

Mini apps for creative work — brainstorm sessions, campaign planning, visual generation, template management. The creative workbench inside the OS.

Mini apps Templates
◌

Task Force

AI agents with specific skills. Assign tasks to AI agents — competitor reports, content drafts, research summaries. Each agent has a defined skill set and brand context.

AI-powered
+

Projects

The container that ties everything together. Each project has its own tasks, chat channel, files, docs, and timeline. Linked to clients from the CRM.

Container
Architecture

Tech stack

Chosen for native performance, developer familiarity, and real-time collaboration capabilities.

Tauri 2 native shell (Rust)
React 19 + Vite
Supabase PostgreSQL + Auth + Realtime + Storage
OpenRouter multi-model AI
LiveKit voice + video
Tiptap + Yjs collaborative docs
Framer Motion animations
Three.js AI sphere
Web Audio API sound effects

Data flow

User input → Brand context injector → OpenRouter API → Claude / GPT / Gemini

Every AI call is enriched with brand context (tone, personas, guidelines) before reaching any model. The response is always on-brand regardless of which model generates it.

Database schema (MVP)

TablePurposeKey fields
teamsTeam containername, logo_url, owner_id
profilesUser profiles + AI prefdisplay_name, avatar, preferred_model
projectsProject containersname, status, deadline, team_id
tasksKanban cardstitle, status, assignee, priority, project_id
messagesChat (Realtime)channel, content, attachments, sender_id
documentsCollaborative docstitle, yjs_state, project_id
brand_assetsBrand vaultname, file_url, type, tags, version
brand_knowledgeAI contextsection, content, team_id
filesFile metadataname, path, mime_type, project_id
notificationsIn-app alertstype, title, read, recipient_id

Platform strategy

PlatformTechnologyTimeline
macOSTauri 2 (WebKit)Phase 1 — launch platform
WindowsTauri 2 (WebView2)Phase 4
iOSTauri 2 Mobile (Safari WebView)Phase 5
AndroidTauri 2 Mobile (Chrome WebView)Phase 5
AI integration

Multi-model AI with brand knowledge

The AI layer is not a chatbot bolted on. It's the intelligence backbone woven into every module.

Voice interface

Press the mic button to speak. Web Speech API transcribes in real-time. The AI responds through a 3D animated sphere (Three.js GLSL shaders) with speech synthesis. Full conversational flow without typing.

Per-user model selection

Each team member picks their preferred model in settings. All models receive the same brand context. Claude for nuanced writing, GPT-4o for speed, Gemini for multimodal. OpenRouter handles routing and billing.

Brand context injection

Before any prompt reaches any model, a system prompt is built from your brand knowledge: tone of voice, personas, positioning, guidelines. Stored in Supabase, editable by the team, injected automatically.

AI agents (Task Force)

Autonomous agents with defined skills — competitor analysis, content drafting, research synthesis. Assign a task to an agent, get a notification when it's done. Each agent inherits the brand context.

Where AI appears

SurfaceActionTrigger
Voice assistantSpeak naturally, get AI responseMic button
Chat @aiAsk questions in any channel@ai mention
Doc editorGenerate drafts in brand voice"Draft with AI" button
Task cardsExpand title into description"AI expand" button
Brand vaultGenerate personas from brief"Generate" button
Brand voice checkReview text against guidelinesRight-click → check
Notifications

Real-time awareness

Three layers of notifications keep the team informed without interrupting flow.

In-app

Bell icon with badge count. Notification center with task assignments, AI completions, chat mentions. Mark as read, click to navigate.

Phase 1

Native desktop

Tauri notification API for background alerts. macOS notification center, dock badge. Sound effects for different event types.

Phase 1

Mobile push

Firebase Cloud Messaging for iOS/Android companion app. Deep links into specific tasks, chats, or docs.

Phase 5
Roadmap

MVP roadmap

Single-user first, multi-user in Phase 3. Each phase delivers a usable product.

1

Foundation

Weeks 1–3 · Single user
  • Tauri 2 + React + Vite scaffold, deployed on macOS
  • Supabase project (EU Frankfurt) with Google OAuth
  • Dashboard with circular scroll menu and start view
  • Tasks module: kanban board, task cards, time tracking
  • Voice AI: mic input → OpenRouter → speech synthesis → sphere animation
  • In-app notification system + native macOS notifications
  • Sound design on all interactions (Web Audio API)
2

Knowledge layer

Weeks 4–6 · Single user
  • Plan module: calendar, governance, project-level kanban
  • Files + Brand module: asset vault, brand core, personas, competitor
  • Create module: mini apps (brainstorm, campaigns, visuals, templates)
  • Markdown documents with Tiptap editor
  • Brand knowledge base — feeds into all AI interactions
  • Supabase Storage for files, images, videos
3

Multi-user + communication

Weeks 7–10
  • Team management: invite links, roles (owner / member)
  • Chat: project channels, DMs, file sharing (Supabase Realtime)
  • Voice + video calls (LiveKit, free 5000 min/month)
  • Task delegation with notifications
  • Collaborative Markdown docs (Yjs CRDT sync)
  • Row-Level Security on all tables for data isolation
  • Task Force: AI agents with skills
4

Intelligence + scale

Weeks 11–14
  • AI agents: autonomous competitor monitoring, content drafting
  • Integrations: Google Drive, Figma API
  • Analytics dashboard: social metrics, revenue, content performance
  • Windows build via Tauri 2 (WebView2)
  • Push notifications via Firebase Cloud Messaging
  • iOS + Android companion app (Tauri 2 Mobile)
LLM context

Context for AI assistants

If you're an LLM helping build Agency OS, here's what you need to know.

Project identity

Project: Agency OS (also referred to as "i7 OS" in UI branding) Author: Sandro Ieva — sandroieva.com — mail@sandroieva.com Type: Native desktop application for creative agencies and micro-studios Stage: Prototype / MVP in active development Repository: github.com/SandroIeva/agency-os-prototype Deployment: Vercel (prototype), Tauri desktop (production target)

Tech stack

Desktop shell: Tauri 2 (Rust) — macOS first, then Windows, then iOS/Android Frontend: React 19 + Vite + Tailwind CSS Backend: Supabase (PostgreSQL + Auth + Realtime + Storage) — EU Frankfurt AI: OpenRouter API (multi-model: Claude, GPT, Gemini, Llama) Animations: Framer Motion (UI) + Three.js GLSL shaders (AI sphere) Voice: Web Speech API (input) + speechSynthesis (output) Sound: Web Audio API (synthesized, no external files) Collaboration: Tiptap + Yjs (CRDT) for real-time doc editing Communication: Supabase Realtime (chat) + LiveKit (voice/video) Styling: Tailwind CSS + Radix UI + Geist font State: Zustand + React Query

Architecture principles

1. SINGLE USER FIRST — build every feature for one user before adding team support 2. BRAND CONTEXT INJECTION — every AI call includes brand voice, personas, guidelines 3. MODEL AGNOSTIC — OpenRouter handles routing; user selects preferred model 4. NATIVE FIRST — use Tauri APIs for notifications, file system, system tray 5. OFFLINE RESILIENT — React Query cache + optimistic updates 6. RESPONSIVE FROM DAY 1 — CSS breakpoints for future mobile companion 7. SUPABASE RLS — user_id on every table from the start, enforce in Phase 3

Design language

Theme: Dark (#0B0B0F background, #111117 surfaces) Font: Geist (sans-serif) — all UI text Accent: Purple-pink gradient (sphere), white (active states) Corners: Large radius (16-28px on cards, 50% on icons) Animations: Framer Motion springs, soft easing [0.22, 0.68, 0.35, 1.0] Sound: Sine wave tones, ascending arpeggios for positive actions Sphere: Three.js with GLSL fragment shader (FBM noise, rose-purple-blue palette) Menu: Circular segmented ring, scroll-to-navigate, orbital card sub-items Icons: Custom SVG icons with rounded stroke style

Module structure

TASKS → Time Tracking, Delegate Tasks CHAT → Team, Clients, Channels, AI Chat, Calls PLAN → Calendar, Governance, Kanban (project level) FILES → Brand Core, Brand Assets, Images, Competitor, Personas, Raw, Links CREATE → Mini Apps (Brainstorm, Campaigns, Visuals, Templates) TASK FORCE → AI Agents, Skillz PROJECTS → Container linking tasks, chat, files, docs per project

Key files

src/App.jsx Main application component (circular menu, voice AI, dashboard) api/chat.js Vercel serverless proxy for Claude/OpenRouter API index.html Entry point with Geist font vite.config.js Vite configuration package.json Dependencies: react, framer-motion, three

Current state

DONE: ✓ Circular scroll menu with segmented ring and orbital sub-items ✓ Start view with greeting and task list ✓ Voice AI: mic → transcription → Claude API → speech synthesis → sphere animation ✓ Three.js GLSL shader sphere with FBM noise color gradients ✓ Framer Motion animations throughout ✓ Web Audio API sound effects on all interactions ✓ Vercel deployment with serverless API proxy TODO (Phase 1): ○ Tauri 2 desktop scaffold (currently web-only prototype) ○ Tasks module with kanban drag-and-drop ○ Supabase schema setup with Google OAuth ○ In-app notification center ○ Time tracking per task ○ OpenRouter integration (replacing direct Claude API)