Asist0 - Agentic Workspace

• 

  architecture-diagram

Asist0 - Agentic Workspace

A voice-first AI assistant built with Google ADK (Agent Development Kit) and the Gemini Live API. Features real-time bidirectional voice streaming, long-term memory across sessions, a file manager with floating editor windows, user-extensible agent skills, image generation/editing, and Firebase-backed storage -- all deployed to Google Cloud.

Architecture

Browser (React)                        Cloud Run (FastAPI + ADK)
+---------------------------+          +------------------------------------+
| File Manager (SVAR)       |          |  WebSocket Handler                 |
| Floating Windows (WinBox) |          |  ADK Runner (run_live)             |
| CodeMirror 6 Editor       |          |                                    |
| PDF Viewer (react-pdf)    |          |  Voice Agent                       |
| WebGL Orb (72px, center)  |--WS----->|  gemini-live-2.5-flash-native-audio|
| Audio I/O (16kHz/24kHz)   |<-WS------|  + PreloadMemoryTool (long-term)   |
| Firestore Realtime Sync   |          |  + SkillToolset (user skills)      |
+---------------------------+          |  + File Tools (13 functions)        |
        |                              |  + Image Tools (generate + edit)   |
        | REST (server fns)            |  + AgentEngineSandboxCodeExecutor  |
        v                              +-------------|----------------------+
+---------------------------+                        |
| TanStack Start (Nitro)    |          +-------------|----------------------+
| Server Functions (proxy)  |--REST--->| Firebase Storage    | Firestore   |
+---------------------------+          | gs://bucket/users/  | users/{uid} |
                                       | {uid}/{path...}     | /files/{id} |
                                       +------------------------------------+

See ARCHITECTURE.md for the full system design.

ADK Features Used

Tech Stack

Project Structure

asisto/
├── asisto_agent/                    # ADK agent package
│   ├── agent.py                     # Voice agent + create_agent() factory
│   └── __init__.py                  #   with PreloadMemoryTool + SkillToolset
├── app/                             # Frontend (TanStack Start + Bun)
│   ├── src/
│   │   ├── lib/
│   │   │   ├── api.ts              # Server functions (file CRUD, workspace, download)
│   │   │   ├── firebase.ts         # Firebase init + Firestore export
│   │   │   ├── useAuth.tsx         # AuthProvider + useAuth hook
│   │   │   ├── useFiles.ts         # Firestore realtime subscription (onSnapshot)
│   │   │   ├── useWorkspace.ts     # Workspace layout save/restore via server fns
│   │   │   ├── useAgentSocket.ts   # Always-on WebSocket (auto-reconnect, exp backoff)
│   │   │   ├── useAudioCapture.ts  # Mic capture (AudioWorklet, 16kHz PCM)
│   │   │   └── useAudioPlayback.ts # Speaker playback (pcm-player, 24kHz)
│   │   ├── components/
│   │   │   ├── Window.tsx          # WinBox.js React wrapper (dynamic import for SSR)
│   │   │   ├── FileViewer.tsx      # CodeMirror + markdown preview + image + PDF viewer
│   │   │   └── Orb.tsx            # WebGL orb (ogl, 72px bottom-center)
│   │   ├── routes/
│   │   │   ├── index.tsx          # Landing / sign-in
│   │   │   └── app/
│   │   │       └── index.tsx      # Main workspace: file manager + floating windows
│   │   └── styles.css             # SVAR vars, Tailwind fixes, WinBox dark theme
│   └── public/
│       └── capture-processor.js   # AudioWorklet (float32 → int16)
├── infra/                          # Pulumi IaC (Python)
│   └── __main__.py                # Cloud Run, Artifact Registry, IAM, domain mappings
├── main.py                         # FastAPI — WebSocket, file REST, workspace REST, auth
├── storage_ops.py                  # Firebase Storage + Firestore CRUD + workspace layout
├── skill_loader.py                 # Reads user skills from Storage, parses SKILL.md
├── agent_tools.py                  # 13 tool closures: file ops + image generation/editing
├── config.yaml                     # Central config (project, region, bucket, engine ID)
├── firebase.json                   # Points to firestore.rules + storage.rules
├── firestore.rules                 # users/{userId}/files + workspace — auth.uid == userId
├── storage.rules                   # users/{userId}/{allPaths=**} — auth.uid == userId
├── Dockerfile                      # Backend container
├── Makefile                        # Build, deploy, dev commands
├── pyproject.toml                  # Python dependencies (uv)
├── ARCHITECTURE.md                 # System design deep dive
├── DEPLOYMENT.md                   # Full deployment guide
└── LICENSE                         # MIT

Quick Start

Prerequisites

• Python 3.13+ with uv
• Bun
• gcloud CLI (authenticated)
• Google Cloud project with Vertex AI + Firebase enabled

Local Development

# Install dependencies
uv sync
cd app && bun install && cd ..

# Configure
cp config.yaml.example config.yaml  # Edit with your project details
cp app/.env.example app/.env

# Deploy Firebase security rules
make deploy-rules

# Run backend (8080) + frontend (3000)
make dev

Make Commands

make dev             Run backend + frontend concurrently
make dev-api         Backend only
make dev-app         Frontend only
make deploy-agent    Deploy agent to Agent Engine
make deploy-infra    Deploy to Cloud Run via Pulumi
make deploy-rules    Deploy Firebase security rules
make deploy-all      Deploy agent + infrastructure
make logs            View backend logs
make status          Check deployment status

Deployment

See DEPLOYMENT.md for the full guide.

License

MIT

Built With

• adk
• firebase
• gemini
• vertex