Rag.NET.Cli 1.0.0

Rag.NET.Cli

A command-line tool for Rag.NET, packaged as a .NET global tool (ragnet) — ingest documents into, and retrieve chunks from, a configured RAG pipeline from the shell, no C# project required.

Install

dotnet tool install -g Rag.NET.Cli

Configure

ragnet wires its pipeline — chat client, embedding generator, vector store — from an appsettings.json next to its working directory, or from environment variables (using __ as the section separator, e.g. RagNet__ChatClient__ApiKey). A sample appsettings.sample.json, showing every supported VectorStore:Kind, ships alongside the installed binaries; copy it to appsettings.json and fill in real values.

{
  "RagNet": {
    "ChatClient": {
      "Endpoint": "https://api.openai.com/v1",
      "ApiKey": "…",
      "Model": "gpt-4o-mini"
    },
    "Embeddings": {
      "Endpoint": "https://api.openai.com/v1",
      "ApiKey": "…",
      "Model": "text-embedding-3-small",
      "VectorDimensions": 1536
    },
    "VectorStore": {
      "Kind": "InMemory | Qdrant | PgVector",
      "Qdrant": { "Host": "…", "Port": 6334, "CollectionName": "…" },
      "PgVector": { "ConnectionString": "…" }
    }
  }
}

This is exactly the configuration seam Rag.NET.Mcp.Tool uses (Rag.NET.Hosting's AddRagNetPipelineFromConfiguration), so the same rules apply: the chat client and embedding generator both go through one OpenAI-compatible endpoint (OpenAI, Azure OpenAI, OpenRouter, Ollama, LM Studio); the vector store is InMemory (the default, zero setup, but every ingested document is lost when the process exits — a warning is logged at startup), Qdrant, or PgVector. A misconfigured setting fails at startup with a message naming both the setting and the configuration key that fixes it, before any command runs.

Need a provider outside that set? Host the Rag.NET.Mcp or Rag.NET library directly in your own application and register whatever you like; this tool covers the bounded set above and nothing wider.

Commands

# Ingest a single file, or every file under a directory (recursively)
ragnet ingest ./document.md
ragnet ingest ./docs [--overwrite]

# Retrieve the chunks a question matches
ragnet query "What is Retrieval-Augmented Generation?" [--top-k 5]

Output goes to stdout as JSON, one object per invocation — meant to be piped to another tool (jq, a script, ...). All diagnostics, warnings, and errors — including the startup validation above and the InMemory warning — go to stderr, never stdout. ingest exits 1 if any file failed (the failures are still listed in the JSON on stdout); query exits 1 if retrieval itself failed.

evaluate — deferred

ragnet evaluate prints an explanation to stderr and exits non-zero; it is not implemented. Rag.NET.Evaluation's evaluators (EmbeddingDistanceEvaluator, LlmJudgeEvaluator) score EvaluationSample instances that already carry a predicted answer — building a working evaluate command means reading a dataset of question/reference pairs in some file format, running each through the pipeline to produce predictions, and choosing which evaluator to run them through. None of that is a thin call onto an existing seam the way ingest/query are: AddRagNetPipelineFromConfiguration registers no IRagEvaluator, and no dataset file format exists anywhere in this repository to parse. Wiring it now would mean inventing that design on the spot rather than reusing one — a half-working evaluate would be worse than an absent one, so it stays absent until that design exists.

Full guide

• MCP server — the same configuration section, for the Rag.NET.Mcp.Tool sibling.