A multi-agent AI workspace platform for orchestrating agent-driven software development. AI agents — developer, architect, designer, QA, DevOps, PM, analyst, and more — collaborate on projects through pluggable LLM backends, with each agent maintaining its own inbox, journal, decision log, and knowledge base.
- Multi-agent orchestration — role-based agents collaborate asynchronously via domain events on a shared project board
- Pluggable LLM executors — swap between Anthropic (Claude), Ollama, LM Studio, GitHub Models, and GitHub Copilot CLI without changing application logic
- Local-first support — run full agent workflows against locally-hosted models with no cloud calls or API keys required
- Progressive code discovery — structured search → slice-read → synthesise pipeline avoids loading full files into context
- Context compaction — automatic transcript compaction with citation preservation to stay within model context budgets
- Model strategy resolution — planning depth, discovery breadth, and tool parallelism are tuned per model capability
- MCP workspace tools — file I/O, directory listing, agent status, journal writing, and git coordination via Model Context Protocol, with audit logging and path validation
- Two UI options — Windows desktop (WinUI 3) and web (ASP.NET Core Razor Components)
| Layer | Technology |
|---|---|
| Runtime | .NET 10, C# 13 |
| Desktop UI | Windows App SDK (WinUI 3) — x86, x64, ARM64 |
| Web UI | ASP.NET Core Razor Components |
| Orchestration | .NET Aspire |
| LLM backends | Anthropic SDK, Claude Code CLI, Ollama, LM Studio, GitHub Models, Copilot CLI |
| Patterns | MVVM Community Toolkit, Result, immutable records |
ai-dev.core/ Core domain entities, feature services, executor contracts, telemetry
ai-dev.core.local/ Local-first orchestration: planning, progressive discovery, compaction, memory
ai-dev.executor.*/ Model-specific executors (Anthropic, Claude, Ollama, LM Studio, etc.)
ai-dev.mcp/ Model Context Protocol implementation and workspace tools
ai-dev-net/ Web UI (ASP.NET Core Razor Components)
ai-dev.ui.winui/ Desktop UI (Windows App SDK / WinUI 3)
ai-dev-net.AppHost/ Aspire app host (dev/deploy entry point)
workspaces/ Agent role templates and workspace definitions
docs/ Architecture design docs and developer guides
Pre-configured agent templates are included for: Developer, Architect, Designer, QA, DevOps, Product Manager, Analyst, Growth Marketer, Guard, and Process Evolution. Each template defines the agent's system prompt, tool access, and interaction patterns for inbox, journal, decisions, and knowledge base.
- .NET 10 SDK
- Windows App SDK (for desktop UI)
- At least one LLM backend:
# Web UI via Aspire
dotnet run --project ai-dev-net.AppHost
# Desktop UI
dotnet run --project ai-dev.ui.winuiSee docs/guides/local-llm-dev-guide.md for full setup instructions when using local models.