DEV Community

Cover image for Chat With Your Files And Database Locally, Open Source
Assistant Engine
Assistant Engine

Posted on

Chat With Your Files And Database Locally, Open Source

I just launched Assistant Engine a C#/.NET 9 native AI interface that runs entirely on your machine, understands your code, talks to your databases, and respects your privacy. No cloud dependency, and no data leaving your device by default.

Assistant Engine is a developer-focused AI tool and UI that lets you run and orchestrate multiple local or remote models via Ollama, do deep C#-aware reasoning, perform Retrieval-Augmented Generation (RAG) over your files, execute PowerShell, and translate natural language into SQL, all from a clean chat interface.

Many Models, One Assistant

This is one of the key design orchestrations, No single model is best at everything. Assistant Engine composes specialized roles into one smart assistant, balancing speed, cost, and expertise. Out of the box you get six roles:

Assistant (the “brain”)
Embedding (vectorizer)
Descriptor (explain before storing)
Correction (fallback models, finetuned in different subjects)
Text-to-SQL
Mini Task (lightweight helpers)

For the devs: You can mix and match models per role, tune temperatures, and scale each role based on the job. You can easily add more roles into your config and pick up the assistant in your own tool via dependency injection.

How Your Data Becomes Searchable:

Describe → Chunk → Vectorize → Store
Assistant Engine’s RAG pipeline turns raw sources into something the assistant can reliably retrieve and cite:

Describe: a descriptor model adds human-friendly summaries and code/schema context.
Chunk: large files are split into ordered segments with rich metadata.
Vectorize: chunks are embedded into dense vectors.
Store: vectors land in local SQLite vector stores

At question time, the assistant searches these stores semantically, pulls back the most relevant chunks, and then answers grounded in your own data.
Files & folders: point the UI at paths to ingest; restrict by extension; choose whether to recurse. There’s also a direct File System Tool for quick list/read/write without ingestion.

Databases: Describe Schemas, Ask Questions, Run Real Queries

Hook up SQL Server, Postgres, or MySQL in Assistant Options → Databases. Flip on Describe Database to generate schema summaries before ingestion — this improves retrieval and Text-to-SQL accuracy. Then just ask questions in plain English; the assistant can search schema, search database, and even execute SQL (best with read-only creds in a sandbox).

Your Stack. Your Data. Your Rules.

Built 100% in .NET 9/C#, it runs natively on your machine with no cloud or Python dependencies. Your files and databases stay private in local vector stores, while its multi-model orchestration gives you full control over speed, cost, and accuracy for every task.

Get Started

GitHub
Website
Documentation (start with RAG Ingestion, Database Access, and Tools)

Top comments (0)