Simple Coding Agent built using Microsoft Agent Framework
| Directory | Purpose |
|---|---|
Agents/ |
Factory methods for planner, builder, and question agents |
Executors/ |
Router, CodePlanner, CodeBuilder, DirectAnswer executors |
Events/ |
Workflow event types and progress data classes |
skills/ |
Domain-specific SKILL.md files and plugin directories |
AgentTools.cs |
Shared agent utility helpers |
- .NET 10 SDK – download
- A local or remote Ollama instance (default:
http://localhost:11434/v1/) running a supported model - Any OpenAI-compatible model API endpoint is also supported
Edit .env to point at your desired model endpoint:
# Model ID — any Hugging Face or Ollama model name
MODEL_ID=qwen3.6:35b-a3b
# API key -- "ollama" for local; your actual key for cloud endpoints
API_KEY=ollama
# OpenAI-compatible API endpoint
OPENAI_ENDPOINT=http://localhost:11434/v1/Tip: Any vendor that implements the OpenAI chat completions API (Groq, Azure OpenAI, Together AI, etc.) works out of the box.
dotnet restore
dotnet builddotnet runYou'll see an interactive prompt:
dotnuxt - .NET Coding Agent (Microsoft Agent Framework)
Model: qwen3.6:35b-a3b
Workspace: /path/to/current/project
Skills directory: /path/to/dotnuxt/skills
Usage: /build <prompt> — Write code, create files, build projects
/ask <prompt> — Answer questions, explain concepts, provide guidance
/skills — List available skills
/plugins — List available plugins
/exit — Quit
> /ask hello, world!
Create a local NuGet package:
dotnet pack -c ReleaseInstall it as a global .NET tool:
dotnet tool install --global dotnuxt --add-source ./nupkgUpdate an existing global install:
dotnet tool update --global dotnuxt --add-source ./nupkgRun it from any project directory:
cd /path/to/another/project
dotnuxtOr explicitly choose a workspace:
dotnuxt --cwd /path/to/another/projectThe installed tool keeps bundled skills/ and .agents/ assets next to the executable, while the workspace remains the directory where you launch dotnuxt or the directory passed to --cwd.
| Variable | Description | Default |
|---|---|---|
MODEL_ID |
Model identifier (e.g. qwen3.6:35b-a3b) |
qwen3.6:35b-a3b |
API_KEY |
API key or "ollama" for local |
ollama |
OPENAI_ENDPOINT |
OpenAI-compatible chat completions URL | http://localhost:11434/v1/ |
dotnuxt loads configuration in this precedence order:
.envin the active workspace%USERPROFILE%\.dotnuxt\.envon Windows, or$HOME/.dotnuxt/.envon macOS/Linux- Existing process environment variables
- Built-in defaults