DenView

A real-time multi-agent office visualizer. Run a task with multiple AI agents and watch them work in an animated office — each agent is a fox character at their own desk, switching between screens and idle activities as their status changes.

Table of Contents

• Overview
• Architecture
• Installation
• Usage

Overview

DenView has three parts:

• Python SDK — instrument your agent code with a few async calls to report task and work state.
• FastAPI server — stores tasks, agents, and work items; serves a state API for the frontend.
• Next.js UI — admin dashboard (tasks, API keys) and an embeddable office view per task.

Each task gets a shareable iframe URL (/view/<task_id>?token=<view_token>) that can be embedded anywhere without authentication.

Architecture

your agent code
    |
    | Python SDK (httpx, async)
    v
FastAPI server  <-->  SQLite (SQLModel)
    |
    | proxied via Next.js route handlers
    v
Next.js UI
    |- /login, /register
    |- /dashboard
    |- /tasks          (table + embedded office view)
    |- /api-keys
    |- /view/[task_id] (embeddable, no auth — view token only)

Auth model

• Admin UI: JWT in httpOnly cookie (/login)
• SDK calls: X-API-Key header
• Embedded view: ?token=<view_token> query param (per-task, read-only)

Office view

• Polls /tasks/<id>/state every 3 seconds
• Each agent has a randomized screen (code, chat, charts, files, terminal, kanban, ...) that rotates every 3–6 seconds while working
• Idle agents cycle through short animations (sleep, stretch, doodle, ...) every 1–2 seconds
• Agents going away (coffee, sofa, window) fade out of their desk

Installation

From PyPI (recommended)

Requirements: Python 3.11+

pip install denview
denview serve --port 8000

On first boot the server creates an admin user with password admin. Change it immediately.

The UI is bundled separately. Run it via Docker (recommended) or clone the repo and run pnpm dev.

Docker (quickstart)

docker compose up

Backend on http://localhost:8004, frontend on http://localhost:3009.

From source

Requirements: Python 3.11+, Node.js 22+, pnpm

git clone https://github.com/your-org/denview
cd denview

# backend
uv sync
denview serve --port 8000

# frontend (separate terminal)
cd ui
pnpm install
pnpm dev

Set the backend URL in ui/.env.local:

BACKEND_URL=http://localhost:8000

Usage

1. Create an API key

Log in at http://localhost:3009, go to API Keys, and create a key.

2. Instrument your agents

pip install denview

from denview.sdk import DenView

dv = DenView(
    api_key="your-api-key",
    backend_url="http://localhost:8004",
    frontend_url="http://localhost:3009",
)

task = await dv.start_task(
    name="My Task",
    description="Optional description",
    agents=[
        {"name": "Alice", "role": "Researcher", "color": "#d95f12"},
        {"name": "Bob",   "role": "Coder",      "color": "#3b82f6"},
    ],
)

print(task.embed_url)  # shareable iframe URL

agent = await task.agent(name="Alice")

# set a markdown note — visible when clicking the agent in the office view
await agent.set_note("""
## Alice — Researcher

Scanning papers and summarizing findings.

- 12 papers in queue
- Flagging contradictions
""")

async with agent.working("refactoring auth module"):
    await do_work()   # marks work failed automatically on exception

await task.finish()
await dv.aclose()

3. Resume an existing task

task = await dv.get_task(task_id=42)
print(task.embed_url)
print(task.data.status)

4. Embed the office view

<iframe src="<task.embed_url>" width="100%" height="500"></iframe>

The embed URL works without login — authenticated by the per-task view_token.

CLI reference

denview serve [--host HOST] [--port PORT] [--reload]