Locally Uncensored / Blog / This Post
Guide October 24, 2026 · 6 min read

How to Use Kimi K3.1 Pro in LU: Chat, Agent and Code, Step by Step

Kimi K3.1 Pro is in LU Cloud since October 24, 2026. This guide shows how to reach it in the Locally Uncensored desktop app, in the browser and on your phone, and how to set up Chat, Agent and Code for it. What the model is and why it cannot run locally yet is in Kimi K3.1 Pro: what it can do.

Before you start

K3.1 Pro runs hosted, in LU Cloud. You need a free LU Cloud account and credits: a plan, or a credit pack from 15 EUR with no subscription. It is the priciest model in the catalog per token, so the Think and effort settings below matter for the bill.

What You Need

Step 1: Sign In to LU Cloud

In the desktop app, open Settings, General, LU Cloud Account and click "Create account", or "Sign in" if you have one. Your browser opens on lu-labs.ai. After the login the page says "Signed in" and the app picks up the session on its own. The account card then shows your plan and your credits.

No credits yet? "View plans" on the same card opens the pricing page. A credit pack is a one time purchase, and its credits do not expire.

Step 2: Switch to Cloud and Pick the Model

Click the Cloud switch in the header next to the model picker. It turns into "Switch to Cloud?"; click it again within six seconds. The picker now lists the hosted models and the composer shows a "Cloud" chip. Open the picker and choose Kimi K3.1 Pro.

You do not need an app update for this. The desktop app reads the list of hosted models from the server, so a new model shows up as soon as it is live. Switching to Cloud also frees your local models from memory; they load again when you switch back.

Step 3: Chat

Type as usual. Two controls next to the message box decide most of the cost:

To ask about a screenshot or a diagram, attach it with the paper clip; the desktop app takes up to five images per message, the browser up to ten. Images need to be at least 64 by 64 pixels.

Step 4: Agent Mode

K3.1 Pro calls tools natively, so it can drive the agent directly. In Cloud mode every hosted model can.

  1. Open a new, empty chat and click the "Agent" toggle above the message box before the first message.
  2. In "Where should the agent work?", choose "Sandbox" for a first try, or "Pick a folder…" for a real project.
  3. Describe the task with the result you want, for example: "Find the three newest releases of this project on GitHub, read their notes and write a summary to notes.md."
  4. Approve the tool calls as they come. You see which tool the model wants to call and with what.

Long agent runs are where K3.1 Pro gains most over K3, and also where it spends most. Every step carries the conversation so far. Repeated parts are billed at a tenth of the input price, but a run of fifty steps still adds up. Start with a task of a few steps and look at the credits on the account card afterwards.

Step 5: The Code Tab

  1. Open the Code tab and pick your project folder with "Select folder...".
  2. Leave the mode on "Ask permissions" at first: commands need your ok, and edits come as a diff before they land.
  3. Pick Kimi K3.1 Pro in the picker and give a coding instruction, such as "Run the tests, find why the date parser test fails and fix it."
  4. Follow the plan bar. Review each diff, then accept or reject it.

The 1M token context lets the model hold a large codebase at once, but you pay for what it reads. Point it at the files that matter instead of asking it to read the whole repository.

In the Browser

  1. Open lu-labs.ai and sign in with the same account.
  2. Open the Studio and pick Kimi K3.1 Pro in the model picker.
  3. Chat, Agent and Code work there too, with the same credits. This is the way on a computer where the desktop app does not run.

On Your Phone

From Your Own Code

Create a key under Cloud API keys on lu-labs.ai (Settings, General, Cloud API Keys in the desktop app links there). The key starts with lu_ and is shown once. Then point any OpenAI compatible client at the endpoint:

curl https://lu-labs.ai/api/inference/v1/chat/completions \
  -H "Authorization: Bearer $LU_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "moonshotai/Kimi-K3.1-Pro",
    "messages": [{"role": "user", "content": "Review this function for edge cases."}]
  }'

Requests through a key always spend credits at the model's rate.

What It Costs

Credits per tokenPer million tokens
Input0.3300,000
Input, repeated part (cache)0.0330,000
Output, thinking included1.51,500,000

A 15 EUR pack carries 450,000 credits. Current plan and pack prices are on the pricing page.

When It Does Not Work

Questions

Do I need a new version of LU for Kimi K3.1 Pro?

No. Version 3.0.7 reads the hosted model list from the server, so K3.1 Pro appears in Cloud mode without an update.

Does it work in Agent mode and the Code tab?

Yes. It calls tools natively, so both run on it directly.

How do I keep the cost down?

Switch Think off or set the effort to Low when a question does not need reasoning, keep agent tasks short, and point the model at the files that matter.

Is Kimi K3 still there?

Yes. Both models stay in LU Cloud, and K3 costs about 5 percent less per token.

Related reading

In the handbook

Locally Uncensored is free, open source and runs on your own machine

Download for Windows Download for Mac (Apple Silicon) Download for Linux

No GPU? LU Cloud runs in a browser, paid from credit packs that start at 15 EUR and need no subscription. See what it costs or open LU Cloud.