The desktop private stack
For individuals who want private AI on their own machine, through apps rather than terminals.
Private AI with nothing to administer: a polished desktop app that browses, downloads and runs open models through a GUI, and a document workspace on top that answers questions from your own files with citations. Both are point-and-click; nothing leaves the machine. Two tools, deliberately: this is the smallest honest version of a private stack.
The stack, step by step
- 01
LM Studio
Run the models: visual model browsing, one-click downloads and a built-in chat, no command line.
Swap options- Jan when open-source auditability and offline-by-default conservatism matter more than the most polished interface See the comparison
- Ollama when you are comfortable in a terminal and want an endpoint that stays on without an app window open See the comparison
LM Studio runs the models and switches on a local API endpoint when applications need to connect; AnythingLLM is exactly such an application, pointing at that endpoint to answer questions from your own files with citations. Runtime below, document workspace above, and nothing leaves the machine.
- 02
AnythingLLM
Ask your documents: point it at LM Studio and chat against your own files, privately, with citations.
Swap options- Open WebUI when the priority is a shared, self-hosted chat interface with accounts rather than a personal document workspace See the comparison
- NotebookLM when the documents are not sensitive and a hosted tool with zero setup is an acceptable trade See the comparison
What it costs
Free software end to end; the only real requirement is a reasonably capable machine, and the only meter is your hardware.
| Tool | Entry tier | What drives cost up |
|---|---|---|
| LM Studio | Free | LM Studio is free for personal use, with the cost being the hardware underneath. Model appetite for memory, not the software, sets the real budget. |
| AnythingLLM | Free | Open source and free to self-host; a hosted cloud version is available. The real cost is the machine and the model behind it. |
Compare the members
Written comparisons between these tools and their nearest substitutes.
Built for
The three tiers of this stack
Ready · this stack
The desktop private stack
For individuals who want private AI on their own machine, through apps rather than terminals.
Competitive
The private and local stack
For teams and individuals whose data cannot leave the building.
World-Class
The production private AI stack
For organisations serving private AI to many users: regulated, air-gapped or simply sovereign by policy.
Common questions
What does this stack actually cost per month?
All two tools here have a genuine free tier, so a working configuration costs nothing while you evaluate it. The 03 COSTS table above breaks down each tool. The meter that climbs here is the hardware underneath: bigger models want more memory, and answer quality follows what your machine can run. Budget for the computer as the main line, since a better machine is the whole upgrade path, and treat each tool as its own decision.
Do I need both tools from day one?
No. LM Studio alone gives you private chat with local models through its own interface, and for general questions that is the whole job. The two steps double as the adoption order, since a model has to run before anything can query it: begin with LM Studio, then add AnythingLLM the moment you want answers drawn from your own files, with citations, rather than the model's general knowledge.
I already use LM Studio. What changes?
Then you already have the runtime half: LM Studio downloads models and runs private chat entirely on your machine, and you keep that exactly as is. What changes is what sits on top. This stack points AnythingLLM at LM Studio's local endpoint, so the same models now answer from your own documents with citations, not just their training. You gain reach over your files without giving up the local guarantee.
Where do these tools overlap, and which wins?
Both give you a chat box, which is the one real overlap. The dividing rule is where the answer comes from. LM Studio's chat wins for general questions and for checking a model quickly: it draws on the model's own knowledge, nothing more. AnythingLLM wins the moment the answer must come from your documents, since it handles ingestion, a local vector store and citations the runtime does not.
When do I outgrow this stack?
Two signals, both about scope, not price. The first is other people: the moment anyone beyond you needs access, a personal desktop app and a single-machine workspace stop fitting, and you want accounts and a shared service. The second is always-on: when the models have to serve reliably in the background rather than through an app window you open. Both point to the competitive tier, the private and local stack.
What can I safely put into these tools?
Almost anything, provided you keep both tools local. Run as described, LM Studio serves models on your machine and AnythingLLM keeps its documents, embeddings and chats on the same box, so sensitive files stay put. The one thing to watch is AnythingLLM's optional cloud model key: turn that on and your text leaves the machine, so leave it off for confidential work.
Before sharing confidential or personal data, check this tool's data-governance and training policies. They differ between providers and can change.
Last checked: July 2026
Where to start
Not sure what to adopt first?
Five quick questions about your job, task and constraints. We'll suggest your top three tools, plus the one to try first.
Tool facts last checked July 2026