No installation, no admin rights, no VPN. Works on Citrix, RDP, VMware VDI and DNS-filtered environments.
VDI Agent is an AI assistant purpose-built for corporate locked-down desktops where traditional software installation is impossible. It runs as a portable .exe overlay on VDI platforms (Citrix, RDP, VMware) and restricted PCs—no admin rights, no persistent installation, no network configuration required.
Deploy in seconds. Works through DNS filters, proxy restrictions, and air-gapped environments. Bring your own AI model or use built-in access to Claude, GPT, Codex.
Keywords: VDI, Citrix, RDP, VMware, locked-down desktops, corporate security, portable, no installation, overlay assistant, DNS bypass, proxy-aware.
| Problem | Solution |
|---|---|
| No software install permission | Portable .exe — zero installation footprint |
| Citrix/RDP limitations | Optimized for VDI screen reading & input injection |
| DNS/proxy filters block cloud AI | Built-in tunnel + DNS bypass + offline mode |
| Session state lost on disconnect | Cloud sync + local cache for persistent context |
| IT security audits | No registry changes, no admin exploitation, no telemetry (optional) |
| Compliance requirements (SOX, HIPAA) | BYOK (bring your own keys), air-gapped, on-prem variants |
| Feature | VDI Agent | GitHub Copilot | MS Copilot | Everywhere (Sylinko) |
|---|---|---|---|---|
| VDI Support (Citrix/RDP/VMware) | ✅ Optimized | ❌ Limited | ❌ No | |
| No Installation | ✅ Portable .exe | ❌ Plugin required | ❌ Web-only | ❌ Installer |
| DNS Bypass / Proxy-Aware | ✅ Yes | ❌ No | ❌ No | ❌ No |
| Multi-LLM Support | ✅ Claude, GPT, Codex, local | ❌ Copilot only | ❌ Proprietary | |
| Portable .exe | ✅ Yes | ❌ No | ❌ No | ❌ No |
| Screen Reading (OCR + Accessibility API) | ✅ Advanced | ❌ No | ||
| Free Tier | ✅ Freemium (200 credits/mo) | ❌ Paid-only | ✅ Limited | |
| Bring Your Own Key (BYOK) | ✅ OpenAI, Anthropic, local | ❌ GitHub account only | ❌ No | ❌ No |
- VDI Platforms: Citrix XenDesktop, Citrix Workspace, VMware Horizon, Microsoft RDS/Hyper-V
- Locked-Down Windows: Domain-joined, AppLocker restricted, no Admin UAC
- Network Restrictions: DNS filter, HTTP/HTTPS proxy, no direct internet access
- Air-Gapped Mode: Local LLM inference, offline cache, manual sync
- Compliance: SOX, HIPAA, PCI-DSS audit trails, no cloud data residency
- Download & Run — Single .exe, zero installation. Works from USB, AppData temp, shared network drive.
- Auto-Detect Environment — Identifies VDI platform (Citrix, RDP, VMware) and applies OS-specific screen reading.
- Screen Capture & OCR — Reads text from locked desktop, respects window boundaries (no pixel-peeping vulnerabilities).
- Route via API or Local — Sends prompts to cloud LLM (Claude, GPT) or local Llama/Mistral inference.
- Inject Response — Types, clicks, pastes result into active window—mimics human user input.
- Sync & Persist — Stores conversation in encrypted local DB, syncs to cloud on reconnect. Resume sessions across VDI disconnects.
- Code Review on Locked Dev Desktop — Paste snippet → AI review → results auto-inserted into IDE
- Security Audit Checklist — OCR compliance form → AI fills fields → submit from locked browser
- Data Analysis in Citrix — Copy table from Excel → AI pivot/chart → paste back to spreadsheet
- Customer Support Scripts — Read ticket → AI drafts response → copy to email (with human review)
- Documentation on Air-Gapped PC — Write offline, sync when network returns, version-tracked
- Frontend: C# WPF (Windows Presentation Foundation) with custom chrome
- Backend: Node.js + TypeScript, async HTTP + gRPC
- LLM Integration: Anthropic API (Claude), OpenAI (GPT), Ollama (local models)
- Screen Reading: Win32 UIAutomation API + custom OCR (EasyOCR, Tesseract)
- Input Injection: SendKeys, clipboard injection, custom IME hooks
- Database: SQLite (local), PostgreSQL (cloud sync)
- Desktop Auth: Device-bound JWT, automatic session refresh
- Deployment: Single .exe (self-contained), GitHub releases, n8n auto-update
| Plan | Free | Starter | Pro | Enterprise |
|---|---|---|---|---|
| Monthly Credits | 200 | 2,000 | 10,000 | Custom |
| AI Models | Claude 3 Haiku | Claude 3 Sonnet | Claude 3 Opus + GPT-4 | BYOK + Local |
| Users | 1 | 1 | 5 | Unlimited |
| VDI Environments | 2 | 5 | 20 | Unlimited |
| Support | Community | Priority | Dedicated | |
| Price | Free | €9/mo | €29/mo | Custom |
BYOK (Bring Your Own Key): Use your own Anthropic/OpenAI API keys. Unlimited credits. €99/mo.
- No Recording — VDI Agent never records screen, audio, or keystroke logs
- Encrypted Local Storage — Conversations encrypted at rest (SQLite + AES-256)
- Zero Telemetry — Optional anonymized usage stats; disabled by default
- BYOK Option — Full data residency in your cloud account (not ours)
- Audit Trail — Every prompt, response, and injection logged for compliance
- No Registry/Admin — Portable .exe has zero system-level persistence
- Open Source Policy — Core screen-reading + injection logic available for audit
- SOC 2 Type II — Annual security certification in progress
Q: Does VDI Agent work on my Citrix XenDesktop? A: Yes. We test on XenDesktop 7.x, Citrix Workspace (cloud), and XenApp. Screen reading uses UIAutomation which Citrix exports. Note: some keyboard injection may require Citrix keyboard sync enabled.
Q: Do I need admin rights to run it? A: No. The .exe runs as your user account (non-admin). No drivers, no registry writes, no UAC prompt. Works in fully locked-down environments (AppLocker, PowerShell restrictions).
Q: What if my corporate DNS filters block API calls? A: VDI Agent auto-detects proxy settings and DNS blocks. If a cloud LLM is unreachable, it silently falls back to offline mode (local Llama/Mistral inference, if enabled). Conversations sync when network returns.
Q: Is this safe to use at work? Could IT flag it? A: VDI Agent leaves zero forensic trace: no .exe installation, no registry keys, no auto-start hooks. However, tell your IT/security team. We provide SOC 2 audit reports and compliance documentation. Most enterprises approve it after review.
Q: Which AI models can I use? A: Built-in: Claude 3 (Haiku/Sonnet/Opus). Bring your own: OpenAI (GPT-4, GPT-3.5), Anthropic, local models via Ollama. Pro tier includes fallback to multiple providers if one is blocked.
- Website: vdiagent.ai
- Download: Releases
- Docs: vdiagent-docs
- Issues & Feedback: GitHub Issues
- Security: SECURITY.md
- .NET 8.0 SDK
- Node.js 20+
- Git
# Clone and restore
git clone https://github.com/ChristianFio/vdiagent-desktop.git
cd vdiagent-desktop
dotnet restore
# Build
dotnet build -c Release
# Run tests
dotnet test
# Publish (self-contained .exe)
dotnet publish -c Release -r win-x64 --self-contained -p:PublishSingleFile=true -p:DebugType=embeddedOutput: VdiAssistant/bin/Release/net8.0-windows/win-x64/publish/VdiAssistant.exe
Contributions welcome. Please:
- Fork & branch:
git checkout -b feature/your-feature - Follow code style: Use EditorConfig (
.editorconfig). Rundotnet formatbefore commit. - Write tests: New features require unit tests. Run
dotnet testlocally. - Commit message:
type(scope): description(e.g.,feat(vdi): add Hyper-V screen reading) - Open PR: Link issues. CI/CD checks must pass.
Code Review: Expect feedback on security, VDI compatibility, and performance. We maintain strict standards for locked-down desktop software.
VDI Agent is dual-licensed:
- Community Edition (GPL-3.0): Free for personal, non-commercial use. Source must remain open.
- Commercial License: For enterprises, closed deployments, and commercial redistribution. Contact sales@vdiagent.ai for pricing.
See LICENSE for full terms.
Last Updated: April 26, 2026
Maintainer: Christian Fiorillo (@ChristianFio)