Features · Local models
Your own models, on your own machine.
If you run models yourself in LM Studio, Ripul finds them and lists them beside Claude, Codex and the rest. The same conversations, the same approvals and the same phone in your pocket, with the model running on hardware you own.
Found without any setting up
Open LM Studio on your Mac and start its server. Ripul Host notices, asks it which models it has, and adds each one to the list you choose from. There is no address to type and nothing to copy across.
Download another model in LM Studio and it appears in Ripul by itself. Close LM Studio and its models leave the list, so you are never offered one that cannot answer.
- Ready ones first
- Models that are already loaded are listed first, so the one that will answer straight away is at the top.
- It says where each one runs
- Every model is labelled with the machine it is on, so the same model on two machines is easy to tell apart.
- It warns you about the wrong kind
- A model that says it cannot use tools is marked, because it may not manage work on files.
A working agent, not only a chat
A local model in Ripul does what the other agents do. It reads your project, changes files, runs things and asks before anything risky, through the same screens. It works through Claude Code, which Ripul points at your model, so Claude Code needs to be installed on the Mac.
Ripul also tells the agent how much your model can hold in mind at once. A long conversation is tidied up before the model runs out of room, at that model's own limit.
A model on another machine
The model does not have to be on the Mac you work at. In Ripul Host's settings, under Coding Agents, add a server by its address and give it a name, and its models join the list. A workstation in the office with a large graphics card is the usual case.
Your Mac makes the calls, so only the Mac has to be able to reach that machine. If the server asks for a key, Ripul Host keeps it on the Mac. It is never sent to your phone.
From your phone
The models on your Mac appear in the list on your iPhone too. Pick one from the sofa or the train, and your Mac does the thinking while you read the answer and say yes or no as usual.
Changing model is a tap inside a running conversation. Begin on a local model, and move to a larger one only when the job turns out to need it.
With a direct connection, nobody else is involved
A local model keeps the thinking on your own hardware. A direct connection keeps the conversation off our servers. They are two halves of one thing: use both, and nothing goes to Ripul and nothing goes to an AI company.
It is a choice for each conversation, so the confidential work can stay on your machines while the rest uses whichever model is best. The Privacy page sets out the three settings side by side.
Good to know
What stays on your machines
With a local model, your questions and the files the agent reads go to the model on your own hardware, not to an AI company. What passes between your phone and your Mac travels the way you chose for that conversation: through Ripul, or direct.
Several agents, one model
A local model slows for everyone if too many agents use it at once. Trainyard can limit how many turns it runs together. See the Trainyard page.
LM Studio is what we test with
Another server can be added by its address if it answers the same kind of request as LM Studio does. We have not tried others yet.
A switch you can turn off
Local Models is a switch in Ripul Host's settings, under Coding Agents. It is on to begin with.
Tools, plans and memories that stay with you, whichever agent made them.