⌕
Ask your documents
Bring in notes, papers, repositories, and folders. Retrieve context locally and answer through a visual FastGPT workflow.
A private workspace for practical AI
Own your AI workspace.
Keep your data local.
Run a useful AI stack on your own machine: private knowledge search, visual RAG workflows, local coding help, and optional LoRA fine-tuning—without usage metering or cloud lock-in.
Runs locally after setupDocker + OllamaWindows and Linux
A focused local stack, assembled from
◉ Ollama◇ FastGPT▣ pgvector✦ BGE Reranker⌁ PEFT + TRL
Built for useful work
One local foundation. Three high-value workflows.
Instead of another chat window, this project gives personal and small-team knowledge work an end-to-end local home.
⌘
Build without sharing code
Use Ollama-backed models and the included OpenCode configuration when your codebase should remain on your machine.
↗
Turn feedback into a model
Collect examples, prepare a dataset, and run the included QLoRA pipeline when you need a specialized local assistant.
Operate and extend safely
More than a chat interface.
Run the stack locally, manage access deliberately, and connect it to tools that keep your data on your machine.
🔐
Authenticated Dashboard
A responsive control plane with viewer, operator, and administrator roles for service actions, user management, and knowledge-base sync.
⬇️
One-click model downloads
Operators and administrators can pull an allowlisted Ollama catalog in the Dashboard while everyone can inspect installed models and job state.
🔌
MCP and agent workflows
Expose local model and reranking tools through stdio MCP, or let Dashboard administrators run step-limited read-only investigations without a terminal.
From question to grounded answer
Simple enough to understand. Powerful enough to extend.
01 / Ingest
Add what you know
Import documents or synchronize selected folders.
02 / Retrieve
Search locally
FastGPT and pgvector find the relevant context.
03 / Rerank
Refine the evidence
The optional BGE service prioritizes stronger matches.
04 / Answer
Use your model
Ollama generates a response in your own workflow.
See it in action
One calm cockpit for your local AI stack.
Service health, model downloads, activity logs, and access control stay visible in one responsive dashboard.
A stack you can inspect
Modern components, no mystery layers.
Each service has a clear role and a documented local endpoint, so you can begin with the defaults and evolve the stack as your needs grow.
✓
FastGPT + pgvectorVisual workflows and vector-backed knowledge bases.
✓
OllamaLocal model inference and embeddings on your own hardware.
✓
BGE rerankerAn optional FastAPI service for improving retrieval order.
✓
LoRA toolingDataset preparation, training, merging, and Ollama export scripts.
Quick start
Run your own workspace.
The setup scripts create a local `.env` with random credentials, pull the recommended Ollama models, and start the services. To use the role-based Dashboard, bootstrap an administrator and run the local control plane after setup. Review the full guide before exposing any service beyond your machine.
8 GB RAM minimumDocker Desktop / DockerPython 3GPU optional
# Linux / WSL — one command setup
git clone https://github.com/caizefan34/local-ai-stack.git
cd local-ai-stack
bash scripts/setup.sh
# Windows — run the guided installer
.\scripts\setup.ps1Ready when you are
Make your AI workspace
yours.
Use it, improve it, and share what you learn. A star helps other builders find a practical path to local AI.