Skip to main content
One installation gives you three entry points. Open a chat and describe your goal in plain language. The fireworks-training compatibility skill carries shared detailed references that configure and debug load. Keep it installed with the three entry skills. Works with Cursor, Codex, OpenCode, Claude Code, and other Agent Skills agents.

​
Install

​
Paste this into your agent

Copy this into the agent you already use. The agent installs the skills. You stay in the chat.
Every training page has a card that copies this prompt. Paste it into your agent. Installing the skills does not need a Fireworks account. The agent asks you to create an account and an API key before any training run.

​
Claude Code

Install the auto-updating cookbook plugin. Here fireworks-training is the plugin name rather than a single skill, and the plugin ships all of the skills above:

​
Cursor

npx skills installs by skill directory, so name each skill you want:

​
Codex

​
OpenCode

​
Other compatible agents

Use the Cursor command and replace cursor with your agent’s id. OpenCode’s id is opencode.
-a '*' installs the skills into every agent the CLI supports, not only the agents on your machine. Pass one agent id unless you want that broader install.
The commands above install skills globally with -g. They do not update automatically. Refresh the global copies with npx --yes skills update -g -y.
Full post-install walkthrough: cookbook skills/GETTING-STARTED.md.

​
Prerequisites

  • Fireworks CLI (firectl) installed. Authenticate with either firectl signin or FIREWORKS_API_KEY.
  • Export FIREWORKS_API_KEY for Training API Python workflows.
  • Prefer a scoped service-account key over a personal admin key for agent use.
If firectl blocks a mutating command inside an AI-agent environment, the skill gives you the exact command to run manually, then resumes read-only monitoring. Use managed training for standard jobs, or the Training API for custom loops on serverless or dedicated infrastructure.

​
Usage data and privacy

Authenticated API calls include fireworks-training-skill/<version> and a random session ID for aggregate product analytics. Prompts and datasets are not collected. Before the first structured question, the skills show a one-line privacy notice. With FIREWORKS_API_KEY set, aggregate journey events may be recorded through supported firectl commands. Say do not track this session to keep interaction telemetry local. Training still works.

​
See also

Getting started (cookbook)

Post-install guide and first run.

Case studies

Runnable notebooks research routes to.

Training Overview

Pick managed training vs the Training API before you install.

Managed Training

Drive the same training infra directly when you know your config.

Training API

Write your own Python training loop on Fireworks GPUs.

CLI reference

Automate managed training with firectl.