Skip to main content
This guide is required before upgrading to Zylon 1.54.0 or any later version. Skipping these steps will cause Triton to fail to boot and may break existing API token access.

​
Overview

This guide walks you through the full upgrade lifecycle in three phases:
  • Pre-upgrade — back up your instance and update the CLI
  • Upgrade — reconfigure and install the new platform version
  • Post-upgrade — migrate API tokens to the new Gateway system (required for ZylonGPT API access)
Plan for approximately 45–60 minutes of total upgrade time, including ~35 minutes of GPU downtime during which inference will not be available. Notify your users before starting.

​
Do I need the token migration?

Only using Workspace endpoints

You’re not affected by the token changes. Complete the platform upgrade steps and skip the post-upgrade section.

Using ZylonGPT API (or planning to)

Tokens generated before v1.54 will only work as Workspace tokens and cannot call /api/gpt/v1/*. You must complete the post-upgrade token migration.

​
Pre-upgrade

1

Back up your instance

Take a full backup of your instance before making any changes. The upgrade only affects the platform layer and drivers — your knowledge base, chats, agent flows, and configuration are not touched — but a backup is strongly recommended.
2

Upgrade the CLI to 2.0.0

Run the CLI self-update to get Zylon CLI 2.0.0:
sudo zylon-cli self-update
Verify the version before continuing:
sudo zylon-cli version
# Expected: 2.0.0 or higher
Do not proceed until the CLI is on version 2.0.0 or higher. Installing v1.53.0+ with an older CLI will result in a broken installation.

​
Upgrade

1

Reconfigure your instance

v1.54.0 requires you to regenerate your config file. Choose one of the following:
  1. Interactive wizard — run sudo zylon-cli config init and follow the prompts.
  2. Silent regeneration — run sudo zylon-cli config init -y to regenerate without prompts.
2

Start the installation

List available versions and install your target version:
sudo zylon-cli list-versions
sudo zylon-cli install <version>
Starting with v1.54.0, each version bundles its own dependencies — including Apt packages, binaries, and drivers — so you get the exact environment required for that version. This release also upgrades to CUDA 13.0.
The installation will intentionally stop with the following message after uninstalling the existing NVIDIA drivers:
The current drivers were uninstalled successfully, please REBOOT the system
and re-run the installation to complete the NVIDIA driver installation
This is expected behavior. Proceed to the next step to reboot before continuing.
3

Reboot the system

After seeing the uninstall message, reboot:
sudo reboot
This is required before the new CUDA 13.0 drivers can be installed.
4

Complete the installation

Once the system is back up, re-run the install command:
sudo zylon-cli install <version>
The installation will now complete successfully with the new CUDA 13.0 drivers.
Going forward, always use sudo zylon-cli install <version> to upgrade. The sudo zylon-cli update command still exists but no longer manages version upgrades as of v1.53.0 — it now acts as sudo zylon-cli sync.

​
Post-upgrade: Migrate API tokens

Skip this section if you are not using the ZylonGPT API (/api/gpt/v1/*).
Tokens generated before v1.54 only work as Workspace tokens. To call ZylonGPT endpoints, you need a Gateway token.
1

Create a Gateway

Log in with your admin@zylon.ai account and create a new Gateway in the Backoffice. Go to Backoffice → Gateways
Only the admin@zylon.ai account has Gateway creation permissions by default. Any operator assigned the Operator role in the Backoffice can also create Gateways.
2

Add members to the Gateway

Once the Gateway is created, add the users who need access as Developer members.
Only users with the Developer role can be added to a Gateway. If a user is not yet a Developer, promote them first in Account Management before adding them here.
3

Generate a new Gateway token

  1. Log in with your user account (not the admin account).
  2. Open Developer Console and select your Gateway.
  3. Click Create API Token, name it, and generate.
  4. Copy the token secret — it is shown only once.
You can now use this token to access all ZylonGPT (/api/gpt/v1/*) endpoints. See the full token management documentation for details.

​
Troubleshooting

The most likely cause is that the CLI was not updated to 2.0.0 before running the install. Verify:
sudo zylon-cli version
sudo kubectl get pods -A | grep triton
If the CLI is older than 2.0.0 and Triton is not showing 1/1 READY, follow the recovery steps in this guide from the beginning, starting with the CLI self-update.
Stop the current install command, manually reinstall the problematic package, then retry:
sudo apt-get install --reinstall libc-bin
sudo zylon-cli install <version>
Reboot the machine and retry. If the issue persists, contact Zylon support. You can also run a faster check by skipping packages and binaries:
sudo zylon-cli install <version> --no-packages --no-binaries
Tokens generated before v1.54 only work as Workspace tokens. If you’re calling /api/gpt/v1/* endpoints, follow the Post-upgrade: Migrate API tokens steps above to create a Gateway and generate a new Gateway token.