How to Deploy a Private Chute on Chutes.ai & Understand Pricing

Last updated: November 11, 2025

1) What is a “Chute”

A chute is your packaged AI application (model + code + runtime) that runs on Chutes.ai’s GPU infrastructure. Chutes handles provisioning, scaling, and routing; you invoke your chute over a simple API. You can keep it private to your account or make it public.

2) Pre-requisites

• A Chutes.ai account with an active plan.
• Chutes CLI installed: pip install chutes
• Authenticate in your terminal: chutes auth login
• Your model/app code (you can start from a template such as a vLLM chute).

3) Deploying a Private Chute (step-by-step)

1. Define your chute (example using a vLLM template):

   from chutes.chute import NodeSelector
   from chutes.chute.template.vllm import build_vllm_chute
   
   chute = build_vllm_chute(
       username="your_username",
       model_name="microsoft/DialoGPT-medium",
       revision="main",
       node_selector=NodeSelector(
           gpu_count=1,
           min_vram_gb_per_gpu=16
       ),
       concurrency=4,
       readme="""
       # My First Chute
       A simple conversational AI.
       """
   )

2. Build the image (packages code, dependencies, and model):

   chutes build my_first_chute:chute --wait

3. Deploy (private by default):

   chutes deploy my_first_chute:chute

   Note: Omit any --public flag to keep the chute private.

4. Review output (example):

   URL: https://your-username-my-first-chute.chutes.ai
   Chute ID: 1234-abcd-efgh
   Status: Running (1/1 instances)

5. Manage & scale (selected commands):

   # List chutes
   chutes chutes list
   
   # Inspect one chute
   chutes chutes get my_first_chute
   
   # Set env vars during deploy
   chutes deploy my_first_chute:chute --env-file .env --wait
   
   # Scale instances
   chutes scale my_first_chute --instances 2

4) Private vs Public Deployment

• Private (default): Only your account (and any users you explicitly share with) can invoke the chute.
• Public: If you deploy with a --public flag (or publish via the dashboard), the chute is visible/usable by others as configured. Billing remains with the owner.

5) Pricing Overview

Chutes.ai pricing typically combines a subscription plan with usage-based compute. Exact pricing and quotas may vary by region, GPU type, and plan.

Compute & inference usage: You are billed for active (“hot”) instances of your chute based on GPU class and runtime. Initial deployments may include a one-time deployment charge (e.g., tied to hourly rate) with ongoing per-hour (or per-second) billing while instances are running, including any configured shutdown delay.

6) Support / FAQ Highlights

• Production readiness: Chutes manages provisioning, scaling, and routing across GPU nodes.
• First-time deploy time: The first build can take several minutes (image + dependencies + model). Code-only updates are typically faster.
• Troubleshooting builds: Use compatible base images, keep Docker contexts small, and verify permissions for any downloaded models or weights.

7) KB Summary

In short: Write or template your model app → chutes build → chutes deploy (omit --public for private) → call it via the provided URL/API → manage scale, env vars, and shutdown settings to control cost. Pricing = plan + usage-based compute.