Manifold Desktop

Open-source multi-provider AI chat client for Windows.

Manifold Desktop is a native Windows desktop client that lets you chat with multiple AI providers through a single unified interface. Instead of juggling separate tabs/apps for ChatGPT, Claude, Gemini, and local models, you use one app with one conversation history, one settings panel, and the ability to switch providers mid-conversation or compare them side-by-side.

Status: Experimental. Feedback and contributions welcome.

Features

• Multi-provider support -- Gemini, OpenAI, Anthropic, Ollama, and any OpenAI-compatible endpoint
• Side-by-side model comparison -- stream responses from two models in parallel
• Integrated terminal -- full ConPTY pseudo-console with ANSI rendering
• MCP client -- Model Context Protocol support with stdio and SSE transports
• Prompt library -- save and reuse system prompts across conversations
• Session management -- save, load, search, rename, export, and import chat sessions
• Markdown export -- export conversations as .md files
• Cost tracking -- per-conversation token usage and cost estimates
• Plugin system -- extend functionality via DLL plugins with virtual host registration
• Secure API key storage -- Windows Credential Manager (DPAPI) encryption
• Onboarding -- guided setup overlay when no API keys are configured
• Window state persistence -- remembers position, size, and maximized state across sessions
• Keyboard shortcuts
  • Ctrl+N -- new chat
  • Ctrl+T -- new terminal
  • Ctrl+W -- close tab
  • Ctrl+F -- search sessions
  • Ctrl+, -- settings
  • Ctrl+K -- focus sidebar search
  • Ctrl+1..9 -- switch tabs

What's New in v0.2.0

• Ollama integration -- chat with local models running on Ollama
• Side-by-side comparison -- compare responses from two providers in parallel
• Prompt library -- save, organize, and reuse system prompts
• Session search -- full-text search across all saved conversations
• Markdown export -- export any conversation to a .md file
• Cost tracking -- per-conversation token counts and cost estimates
• Session export/import -- portable JSON-based session backup and restore
• Auto-update check -- notifies you when a new version is available
• Onboarding flow -- first-run overlay guides API key configuration
• Window state persistence -- remembers window position, size, and maximized state
• Inline session rename -- double-click a session name to rename it
• Token estimates -- live token count below the input textarea
• Loading indicators -- visual feedback during AI response streaming
• Toast notifications -- non-intrusive status messages
• Confirmation dialogs -- safe tab close and session delete with confirmation prompts
• Accessibility -- ARIA attributes, focus-visible indicators, and improved keyboard navigation
• Error recovery -- inline error display with one-click retry

See the full changelog below.

Quick Start

Prerequisites

• Windows 10 version 1903 or later (x64)
• Visual Studio 2022 or later with:
  • Desktop development with C++ workload
  • Universal Windows Platform development workload
  • Windows 10 SDK (10.0.19041.0 or later)
• Windows App SDK 1.8

Clone and build

git clone https://github.com/gregorik/Manifold-Desktop.git
cd Manifold-Desktop

Restore NuGet packages and build:

msbuild ManifoldDesktop.vcxproj /p:Configuration=Release /p:Platform=x64 /restore

Run the output from x64\Release\ManifoldDesktop\.

Build from Source

Debug build

msbuild ManifoldDesktop.vcxproj /p:Configuration=Debug /p:Platform=x64 /restore

Release build

msbuild ManifoldDesktop.vcxproj /p:Configuration=Release /p:Platform=x64 /restore

The /restore flag triggers NuGet package restore automatically. If you prefer to restore separately:

nuget restore ManifoldDesktop.sln
msbuild ManifoldDesktop.vcxproj /p:Configuration=Release /p:Platform=x64

You can also open ManifoldDesktop.sln in Visual Studio, which handles NuGet restore on build.

Architecture

Manifold Desktop is a two-layer application:

• C++/WinRT backend -- manages providers, terminal sessions, MCP connections, file I/O, and credential storage
• WebView2 frontend -- vanilla JavaScript (ES6 modules) with no framework or build step

The two layers communicate via a JSON message bridge (postMessage / OnWebMessage). Each provider implements the IProvider interface and is registered at startup through ProviderRegistry.

See docs/architecture.md for a detailed breakdown.

Extending

Adding a Provider

1. Create a class that implements Manifold::Core::IProvider (see Manifold.Core/Providers/IProvider.h).
2. Implement all virtual methods: Id(), DisplayName(), ListModels(), SendChat(), StreamChat(), and ValidateKey().
3. Register an instance in MainWindow's constructor alongside the existing providers:

m_providerRegistry.AddProvider(std::make_unique<YourProvider>());

4. Rebuild. The new provider will appear in the frontend's provider dropdown automatically.

Adding an MCP Server

1. Open Settings (Ctrl+,) and navigate to the MCP section.
2. Add a server with either:
   • Stdio transport -- provide the command and arguments (e.g., npx -y @modelcontextprotocol/server-filesystem /path)
   • SSE transport -- provide the server URL (e.g., http://localhost:8080/sse)
3. Enable the server. Manifold will connect and discover available tools, which are then offered to AI providers that support tool use.

Changelog

v0.2.0 (2026-04-03)

New features:

• Ollama provider integration for local model support
• Side-by-side model comparison mode
• Prompt library with save/load/organize
• Full-text session search
• Markdown conversation export
• Per-conversation cost tracking and token estimates
• Session export/import (JSON format)
• Auto-update check on launch
• First-run onboarding overlay
• Window position/size/maximized state persistence
• Inline session rename (double-click)
• Live token estimate below input textarea
• Loading indicators during streaming
• Toast notification system
• Confirmation dialogs for destructive actions
• Accessibility improvements (ARIA, focus-visible, keyboard nav)
• Inline error display with retry

Bug fixes:

• Terminal theme variable consistency
• Tab transition animations
• Disabled button state styling

v0.1.0 (2026-01-27)

• Initial release
• Multi-provider support (Gemini, OpenAI, Anthropic, OpenAI-compatible)
• Integrated ConPTY terminal
• MCP client (stdio + SSE)
• Plugin system
• Secure credential storage
• Session save/load

Contributing

See docs/contributing.md for setup instructions, code style, and PR guidelines.

License

MIT. Code by Andras Gregori @ GregOrigin.