Skip to main content
Create an account, connect your cloud account, and start running your workflows in minutes.

​
1. Create an account

Before you get started, make sure to create an account at app.komodo.io.

​
2. Install Komodo CLI and authenticate

Install the CLI: 
pip install komo
Head to the Settings page and copy your API key. Now login and provide your API key when prompted: 
komo login

​
3. Launch a machine

Let’s spin up a development machine Copy the following YAML into a dev.yaml file: 
resources:
  # 1x A10 GPU
  accelerators: A10:1

envs:
  # set the environment variable MY_ENV_VAR to my-value
  MY_ENV_VAR: my-value

# Copy the contents of the current directory onto the remote machine
workdir: .

# Typical use: pip install -r requirements.txt
# Invoked under the workdir (ie. can use its files)
setup: |
  echo "Running setup operations"

# Start a notebook on the machine
notebook: true
And then simply run: 
komo machine launch dev.yaml --name dev
Once the machine is launched, you can open the notebook from dashboard. You can also connect to the machine with SSH or open it in VSCode using the CLI.

​
3. Launch a job - Hello World!

Let’s define our very first job, a simple Hello World! program. Copy the following YAML into a hello.yaml file: 
resources:
  # 1x A100 GPU
  accelerators: A100:1

envs:
  # set the environment variable MY_ENV_VAR to my-value
  MY_ENV_VAR: my-value

# Copy the contents of the current directory onto the remote machine
workdir: .

# Typical use: pip install -r requirements.txt
# Invoked under the workdir (ie. can use its files)
setup: |
  echo "Running setup operations"

# Typical use: python train.py
# Invoked under the workdir (ie. can use its files)
run: |
  echo 'Hello World!'
You’ll notice this is the exact same YAML config as the previous one above, with an additional run section. This section defines how to run your job. To launch this job, simply run: 
komo job launch hello.yaml
And that’s it! The logs from the job will be streamed by the CLI in real-time, and you can also access them from the web dashboard. Once the job finishes, the remote instance will be terminated.

​
4. Launch a service

Now let’s define our very first service, a simple python HTTP server. Copy the follwing YAML into a http-service.yaml file: 
resources:
  # 1x A10 GPU
  accelerators: A10:1
  # Open port 8000
  ports: [8000]

envs:
  # set the environment variable MY_ENV_VAR to my-value
  MY_ENV_VAR: my-value

# Copy the contents of the current directory onto the remote machine
workdir: .

# Typical use: pip install -r requirements.txt
# Invoked under the workdir (ie. can use its files)
setup: |
  echo "Running setup operations"

# Typical uses:
#    torchserve ..
#    python -u -m vllm.entrypoints.openai.api_server ...
# Invoked under the workdir (ie. can use its files)
run: |
  python -m http.server --port 8000

service:
  replica_policy:
    min_replicas: 1
    max_replicas: 5
    # When the average QPS (queries per second) per replica goes above this number,
    # Komodo will dynamically scale up the number of replicas running your service,
    # up to a maximum of max_replicas.
    # Similarly, when the average QPS per replica goes below this number, Komodo will
    # scale down the number of replicas running your service, down to a minimum of min_replicas.
    target_qps_per_replica: 10
  
  readiness_probe:
    # this is the endpoint path within your service that Komodo uses to check if your service is running
    path: /
Again, you’ll notice this is the exact same YAML config as the previous one above, but this time, in our run section, we start our service on port 8000 (and expose port 8000 in the resources section), and the service sections defines the parameters that are used to determine how to scale your service. To launch this service, simply run: 
komo service launch http-service.yaml --name simple-http-service