Skip to content

Frequently Asked Questions

Quick answers to common questions about DeskAgent.


General

What is DeskAgent?

DeskAgent is an AI-powered desktop assistant that automates office tasks. It connects to your tools (email, invoicing, support tickets) and handles repetitive work like organizing inboxes, creating quotes, and processing invoices.

How does it work?

  1. You click an agent tile or type a request in the chat
  2. The AI understands what you need and plans the steps
  3. DeskAgent executes the actions using your connected tools
  4. You review the results before anything is sent

Is my data sent to the cloud?

Only AI prompts are sent to the AI provider (Anthropic/Google). Your actual emails, documents, and business data stay on your local machine. The AI sees the content it needs to process, but:

  • Files are not uploaded to any cloud storage
  • Sensitive data can be anonymized before AI processing
  • You can use local AI models (Ollama) for complete privacy

Which AI providers are supported?

Provider Best For Privacy
Claude (Anthropic) Complex tasks, customer-facing content Cloud
Gemini (Google) Daily operations, cost-conscious users Cloud
Ollama (local) Privacy-first, offline work 100% local

Email

Does it work with my email provider?

DeskAgent uses Microsoft Outlook. It works with any email account configured in Outlook:

  • Microsoft 365 / Office 365
  • Exchange Server
  • Outlook.com
  • Gmail (via Outlook)
  • IMAP/POP accounts (via Outlook)

Does Outlook need to be open?

Yes. DeskAgent communicates with Outlook through Windows COM automation, which requires Outlook to be running.

Can DeskAgent send emails automatically?

By default, DeskAgent creates drafts, not sent emails. You always review before sending. This is a safety feature. Automatic sending can be enabled for specific agents if needed.


Billing & Invoicing

Which billing systems are supported?

  • Billomat - German online invoicing
  • Lexware Office - German accounting software

Both can be used simultaneously.

Can I add other billing systems?

Yes, DeskAgent is extensible. New billing integrations can be added by creating an MCP server. See Creating Agents for the technical approach.

What about GDPR?

DeskAgent is designed for EU compliance: - Data stays on your local machine - Personal data can be anonymized before AI processing - You control what data is processed - No data is stored in external systems (beyond what the AI providers state in their policies)


Installation & Updates

The installer fails - what should I do?

  1. Run as Administrator - Right-click → "Run as administrator"
  2. Check antivirus - Temporarily disable Windows Defender or whitelist the installer
  3. Check internet - The installer downloads dependencies
  4. Check logs - Look at setup.log in the install directory

How do I update DeskAgent?

Run the update script:

.\deskagent\update.bat

This pulls the latest code and updates dependencies.

Can I run DeskAgent on Mac or Linux?

Currently Windows only, because Outlook COM automation is Windows-specific. A cross-platform version using Microsoft Graph API is on the roadmap.


Configuration

Where are my API keys stored?

In the config/ folder: - config/backends.json - AI provider keys (Anthropic, Google) - config/apis.json - Service credentials (Billomat, UserEcho, etc.)

These files are in .gitignore and never committed to version control.

Can I use multiple AI backends?

Yes! Configure different backends in backends.json and assign them to specific agents:

// In agents.json
"daily_check": {
  "ai": "gemini"      // Use Gemini for routine tasks
},
"create_offer": {
  "ai": "claude_sdk"  // Use Claude for customer-facing content
}

How do I switch between backends?

Either: 1. Change the default in backends.json, or 2. Assign specific backends to specific agents in agents.json


Agents

How do I create a custom agent?

  1. Create a markdown file in agents/ describing the workflow
  2. Add configuration in config/agents.json
  3. Restart DeskAgent

See Creating Agents for a complete guide.

An agent failed - how do I debug?

  1. Check the system log: workspace/.logs/system.log
  2. Review the agent log: workspace/.logs/agent_latest.txt
  3. Try step-by-step: Run each step manually in chat mode

Can agents run automatically?

Yes, three ways:

Method How Best For
Email Watcher Triggers when matching emails arrive Invoice processing
Teams Watcher Triggers on Teams messages Team notifications
Task Scheduler Windows Task Scheduler runs at set times Daily Check

Security

Is DeskAgent secure?

DeskAgent implements defense-in-depth:

Layer Protection
MCP filtering Agents only access allowed tools
Tool whitelist Restrict to specific actions
Path restrictions Limit file system access
PII anonymization Mask personal data before AI processing
Prompt injection protection Sanitize external content

Can I restrict what an agent can do?

Yes, using allowed_mcp and allowed_tools:

"my_agent": {
  "allowed_mcp": "outlook|clipboard",
  "allowed_tools": ["get_selected_email", "set_clipboard"]
}

This agent can only read emails and use the clipboard - nothing else.

See Security Guide for details.


Troubleshooting

DeskAgent won't start

Symptom Solution
"Port already in use" Run .\deskagent\stop.bat first
No window appears Check workspace/.logs/system.log for errors
Python not found Ensure Python is installed or use the installer

Outlook tools don't work

Symptom Solution
"Outlook not available" Make sure Outlook is open (not just in system tray)
"No email selected" Click once on an email in Outlook
COM errors Delete gen_py/ folder and restart

AI responses are slow

Cause Solution
Slow internet Check your connection
Complex request Break into smaller steps
Large context Use knowledge patterns to limit context

For more solutions, check the system log at workspace/.logs/system.log.


Cost

How much does it cost?

DeskAgent itself is free and open source. You pay for AI API usage:

Provider Input Cost Output Cost
Claude Sonnet $3/1M tokens $15/1M tokens
Gemini Pro $1.25/1M tokens $10/1M tokens
Ollama (local) Free Free

How can I reduce AI costs?

  1. Use Gemini for routine tasks - 60% cheaper than Claude
  2. Use local models - Free but less capable
  3. Limit context - Use knowledge patterns in agent config
  4. Reduce tool descriptions - Use allowed_mcp to load only needed tools

Getting Help

Where can I get support?

Where can I report bugs?

Open an issue on GitHub or post in the forum with: 1. What you tried to do 2. What happened instead 3. Contents of workspace/.logs/system.log