Generellem 1.1.0

.NET 10.0 This package targets .NET 10.0. The package is compatible with this framework or higher. 
dotnet add package Generellem --version 1.1.0
                    


                    

                
NuGet\Install-Package Generellem -Version 1.1.0
This command is intended to be used within the Package Manager Console in Visual Studio, as it uses the NuGet module's version of Install-Package. 
<PackageReference Include="Generellem" Version="1.1.0" />
For projects that support PackageReference, copy this XML node into the project file to reference the package. 
<PackageVersion Include="Generellem" Version="1.1.0" />
                    


                            Directory.Packages.props
                        

                    

                
<PackageReference Include="Generellem" />
                    


                            Project file
For projects that support Central Package Management (CPM), copy this XML node into the solution Directory.Packages.props file to version the package. 
paket add Generellem --version 1.1.0
                    


                    

                
#r "nuget: Generellem, 1.1.0"
#r directive can be used in F# Interactive and Polyglot Notebooks. Copy this into the interactive tool or source code of the script to reference the package. 
#:package Generellem@1.1.0
#:package directive can be used in C# file-based apps starting in .NET 10 preview 4. Copy this into a .cs file before any lines of code to reference the package. 
#addin nuget:?package=Generellem&version=1.1.0
                    


                            Install as a Cake Addin
                        

                    

                
#tool nuget:?package=Generellem&version=1.1.0
                    


                            Install as a Cake Tool

Generellem Logo

Generellem

Generellem is an opinionated .NET framework for Retrieval Augmented Generation (RAG). Point it at your documents and ask questions.

Generellem 1.1.0 targets .NET 10 and runs on Linux, macOS, and Windows. The default console profile needs no Azure account, no API keys, and no Docker.

Try it

Clone the repo and run the console demo. The InMemory profile is the default.

git clone https://github.com/JoeMayo/generellem
cd generellem/GenerellemConsole
dotnet run

The first run ingests documents from this repository. bin, obj, and .git are skipped. It also requests the project wiki home page, so a network connection helps. When that request fails, local documents are still ingested.

You will see ingestion progress, then a banner that says Profile: InMemory. At the prompt, try:

generellem> How do I contribute to Generellem?

The answer comes from files such as CONTRIBUTING.md, followed by a Sources list. InMemory and Sqlite do not call a chat model. They return extractive answers from the retrieved text. For a generated answer, use the Ollama profile below.

Type quit to exit.

Profiles

Profile Command What you need
InMemory (default) dotnet run Nothing. Extractive answers, kept in memory for the process.
Sqlite dotnet run -- --profile Sqlite Nothing. Same extractive answers. Vectors persist in SQLite.
Ollama dotnet run -- --profile Ollama Ollama, then ollama pull llama3.2 and ollama pull nomic-embed-text.
Azure dotnet run -- --profile Azure Azure OpenAI plus Qdrant. Set Generellem:UseAzureSearch to true to use Azure AI Search instead.

Run ingest, questions, or both:

dotnet run -- --mode ingest
dotnet run -- --mode query
dotnet run -- --mode both

Generellem:RunMode in config is the same switch. --mode overrides it. An unknown mode falls back to Both.

Optional Docker services for Qdrant and Ollama:

docker compose up -d
docker compose exec ollama ollama pull llama3.2
docker compose exec ollama ollama pull nomic-embed-text

Do not run a second Ollama on port 11434 at the same time. The comments in docker-compose.yml describe that setup.

Add it to an app

services.AddGenerellem(GenerellemProfile.InMemory);
// or GenerellemProfile.Sqlite, Ollama, or Azure

Ingestion and query code talk to ISearchService. The profile picks the implementation.

What you get

  • Hybrid search on every backend. Azure AI Search combines keyword and vector queries. Qdrant adds a lexical sparse vector and fuses results with reciprocal rank fusion, and falls back to dense search when sparse search is unavailable. SQLite uses FTS5. InMemory fuses keyword overlap with cosine similarity. A blank query stays vector-only.
  • Reranking before the prompt is built. Generellem:RerankMode is Lexical (default), Llm, or None. Lexical keeps the original order on ties, so the default does not add a model call.
  • Citations. Retrieved text is wrapped as untrusted context. The answer is followed by a distinct Sources list.
  • Streaming. AskStreamAsync yields tokens, then the citation suffix. Both demos print the stream. AskAsync stays non-streaming.
  • Images. .jpg, .jpeg, .png, .gif, and .bmp are ingested from embedded text (PNG text chunks and JPEG comments) and the file name. Set EnableImageVision and VisionModel for an Ollama or Azure vision pass. Vision is off by default.
  • sqlite-vec. The Sqlite profile uses the bundled sqlite-vec 0.1.9 index when the host can load it (osx-arm64, osx-x64, linux-x64, win-x64). Otherwise search scans with cosine similarity.
  • Releases. The package version is 1.1.0. Push a vMAJOR.MINOR.PATCH tag to pack Generellem and publish a GitHub release.

Configure

Local options live in GenerellemConsole/appsettings.json and GenerellemConsole/appsettings.Local.json, under the Generellem section.

Setting Default Purpose
ChunkSize 5000 Characters per chunk. Text breaks on paragraph, then sentence, then word.
ChunkOverlap 100 Overlap between chunks. Must be less than ChunkSize.
TopK 3 How many chunks search returns.
Temperature 0 Chat sampling temperature.
SummarizeUserIntent true Rewrites the question before search. Ollama and Azure use the chat model. InMemory and Sqlite keep the question. Set false to skip the step.
EmbeddingBatchSize 16 How many chunks are embedded per ingestion call.
EmbeddingDimensions by profile 64 for InMemory and Sqlite, 768 for Ollama, 1536 for Azure. Override this when your embedding model uses another size.
RunMode Both Both, Ingest, or Query.
RerankMode Lexical Lexical, Llm, or None.
EnableImageVision false When true, VisionModel is required.
VisionModel empty Vision model name for Ollama or Azure.
UseAzureSearch false Azure profile only. false uses Qdrant. true uses Azure AI Search.
OllamaEndpoint http://localhost:11434 Ollama base URL.
OllamaChatModel llama3.2 Chat model.
OllamaEmbeddingModel nomic-embed-text Embedding model.
OllamaUseSqliteSearch false On the Ollama profile, store vectors in SQLite instead of memory.

On each ingest, the console copies FileSystem.json and Website.json into the AppData folder Generellem/ConsoleDemo. FileSystem.json points at the repository root. Edit those files in GenerellemConsole before you run.

Azure keys

The Azure profile reads user secrets in Debug builds. From GenerellemConsole:

dotnet user-secrets set GenerellemAzOpenAIApiKey "<key>"
dotnet user-secrets set GenerellemAzOpenAIEndpointName "<endpoint>"
dotnet user-secrets set GenerellemAzOpenAIDeploymentName "<chat-deployment>"
dotnet user-secrets set GenerellemAzOpenAIEmbeddingName "<embedding-deployment>"

Qdrant, which is the default Azure search backend:

dotnet user-secrets set GenerellemQdrantEndpoint "<endpoint>"
dotnet user-secrets set GenerellemQdrantApiKey "<key>"
dotnet user-secrets set GenerellemQdrantCollection "<collection>"

Azure AI Search, when UseAzureSearch is true:

dotnet user-secrets set GenerellemAzSearchServiceEndpoint "<endpoint>"
dotnet user-secrets set GenerellemAzSearchServiceAdminApiKey "<key>"
dotnet user-secrets set GenerellemAzSearchServiceIndex "<index>"

Deployment names are the names you chose in Azure OpenAI. The embedding deployment's dimensions must match EmbeddingDimensions (1536 for text-embedding-ada-002). For the Qdrant service in docker-compose.yml, the endpoint is http://localhost:6334. Every Azure key above must be non-blank. InMemory and Sqlite need none. Ollama needs an endpoint and both model names.

If you already have a search index from an older build, ingest again so the new chunk boundaries and index fields are stored.

Samples

  • GenerellemConsole is the demo to run first.
  • MSGraphDemo reads OneDrive through Microsoft Graph. It needs a Graph client. It is not part of the quick start.

The Documentation folder is sample content the demo can ingest. It is not the getting-started guide.

Contributing

Contributions are welcome. Read CONTRIBUTING.md. Bugs and ideas go in Issues.

Joe Mayo: X · LinkedIn · BlueSky

Product Compatible and additional computed target framework versions.
.NET net10.0 is compatible.  net10.0-android was computed.  net10.0-browser was computed.  net10.0-ios was computed.  net10.0-maccatalyst was computed.  net10.0-macos was computed.  net10.0-tvos was computed.  net10.0-windows was computed. 
Compatible target framework(s)
Included target framework(s) (in package)
Learn more about Target Frameworks and .NET Standard.

NuGet packages

This package is not used by any NuGet packages.

GitHub repositories

This package is not used by any popular GitHub repositories.

Version Downloads Last Updated
1.1.0 95 10/30/2026
1.0.0.9-beta2 184 7/29/2024
1.0.0.2-beta1 180 2/21/2024

1.1.0 adds hybrid search, reranking, image text and vision ingestion, streamed answers, sqlite-vec indexing, and the .NET 10 target. Package versions follow semantic versioning. Push a vMAJOR.MINOR.PATCH tag to publish a GitHub release.

1.0.0 was the first stable release, with file system, web, and OneDrive ingestion.