cowork-dash 0.1.4

AI Agent Web Interface with Filesystem and Canvas Visualization

Cowork Dash

A web interface for AI agent interactions with filesystem workspace, canvas visualization, and real-time streaming.

Features

• AI Agent Chat: Real-time streaming with thinking process and task progress
• File Browser: Interactive file tree with lazy loading
• Canvas: Visualize DataFrames, Plotly/Matplotlib charts, Mermaid diagrams, images
• Flexible Configuration: Environment variables, CLI args, or config file

Quick Start

Installation

# Install via pip (includes DeepAgents)
pip install cowork-dash

# Or run directly with uvx (no installation needed)
export ANTHROPIC_API_KEY="your_anthropic_api_key"
uvx cowork-dash run --workspace ~/my-workspace

Run

After setting up your agent (optional), run the app. You can also use the default agent by setting ANTHROPIC_API_KEY environment variable.

# Run with defaults (current directory as workspace, no agent)
export ANTHROPIC_API_KEY="your_anthropic_api_key"
cowork-dash run

# Run with workspace
cowork-dash run --workspace ~/my-workspace

# Run with custom agent (optional)
cowork-dash run --agent my_agent.py:agent

# Using uvx (one-off execution)
uvx cowork-dash run --workspace ~/my-workspace --port 8080

Open browser to http://localhost:8050

Configuration

Priority (highest to lowest)

1. CLI Arguments - --workspace, --port, etc.
2. Environment Variables - DEEPAGENT_*
3. Config File - config.py defaults

Environment Variables (optional)

export DEEPAGENT_SPEC=my_agent.py:agent         # Set any Langgraph agent
export DEEPAGENT_WORKSPACE_ROOT=/path/to/workspace
export DEEPAGENT_PORT=9000                      # optional (default: 8050)
export DEEPAGENT_HOST=0.0.0.0                   # optional (default: localhost)
export DEEPAGENT_DEBUG=true                     # optional (default: false)
export DEEPAGENT_APP_TITLE="My App"             # optional
export DEEPAGENT_APP_SUBTITLE="Subtitle"        # optional

cowork-dash run

CLI Options (all optional)

cowork-dash run [OPTIONS]

  --workspace PATH        Workspace directory (default: current directory)
  --agent PATH:OBJECT     Agent spec (default: none, manual mode)
  --port PORT            Server port (default: 8050)
  --host HOST            Server host (default: localhost)
  --debug                Enable debug mode
  --title TITLE          App title (default: "Cowork Dash")
  --subtitle TEXT        App subtitle (default: "AI-Powered Workspace")

Python API

from cowork_dash import run_app

# Option 1: Pass agent instance directly (recommended)
from my_agent import MyAgent
agent = MyAgent()
run_app(agent, workspace="~/my-workspace")

# Option 2: Use agent spec
run_app(agent_spec="my_agent.py:agent", workspace="~/my-workspace")

# Option 3: Manual mode (no agent)
run_app(workspace="~/my-workspace", port=8080, debug=True)

Agent Integration

Workspace Access

Cowork Dash sets DEEPAGENT_WORKSPACE_ROOT environment variable for your agent:

import os
from pathlib import Path

# In your agent code
workspace = Path(os.getenv('DEEPAGENT_WORKSPACE_ROOT', './'))

# Read/write files in workspace
config_file = workspace / "config.json"

Agent Specification

Load agents using path:object format:

# Load from Python file
cowork-dash run --agent agent.py:my_agent

# Absolute path
cowork-dash run --agent /path/to/agent.py:agent_instance

Agent Requirements

Your agent must implement:

• Streaming: agent.stream(input, stream_mode="updates")
• Message format: {"messages": [{"role": "user", "content": "..."}]}
• Workspace access (optional): Read DEEPAGENT_WORKSPACE_ROOT env var

Example Agent Setup

# my_agent.py
import os
from deepagents import create_deep_agent
from deepagents.backends.filesystem import FileSystemBackend

backend = FileSystemBackend(root=os.getenv('DEEPAGENT_WORKSPACE_ROOT', './'))
my_agent = create_deep_agent(..., backend=backend)

Then run: cowork-dash run --agent my_agent.py:my_agent

Canvas

The canvas displays agent-created visualizations:

• DataFrames: HTML tables
• Charts: Plotly, Matplotlib
• Images: PNG, JPG, etc.
• Diagrams: Mermaid (flowcharts, sequence diagrams)
• Markdown: Text and notes

Content auto-saves to canvas.md and can be exported or cleared.

Development

# Install from source
git clone https://github.com/dkedar7/cowork-dash.git
cd cowork-dash
pip install -e ".[dev]"

# Run tests
pytest

# Build package
python -m build

Requirements

• Python 3.11+
• Dash 2.0+
• dash-mantine-components
• pandas, plotly, matplotlib, Pillow
• python-dotenv
• deepagents

Links

• PyPI: https://pypi.org/project/cowork-dash/
• GitHub: https://github.com/dkedar7/cowork-dash
• Issues: https://github.com/dkedar7/cowork-dash/issues