Ways to use distil labs

There are four ways to use the platform. Three of them run jobs: the agent skill, which runs the CLI for you, the CLI itself, and the REST API. They create the same entities, accept the same config overrides, and identify each entity by the same UUID, so a project can move between them freely. The fourth, the dashboard, only reads.

Way	What you can do	Best for
Agent skill	Everything the CLI does, run for you by your coding agent	The recommended way to build a model, your first one included
CLI	Everything: prepare data, run every job, read metrics, deploy	Direct control over each command. The agent skill runs on top of it
REST API	Everything the CLI does, over HTTP	Python pipelines and your own tooling
Dashboard	Read your models, datasets and traces, with their statuses and metrics	Watching a run without a terminal

The CLI

The distil CLI runs every job on the platform, and the agent skill runs it for you. The commands on this site use it. Install it with the install script:

curl -fsSL https://cli-assets.distillabs.ai/install.sh | sh

The script writes one binary to ~/.local/bin/distil. It changes nothing else, and it edits no shell configuration file.

Check that it answers:

distil --version

If your shell cannot find the binary, add ~/.local/bin to your PATH and try again.

Then create your account with distil signup: Account and authentication.

Update the CLI

Update in place:

distil update

Check your version with distil --version.

Supported platforms

Platform	Supported
Linux (x86_64)	Yes
Linux (arm64)	Yes
macOS (Intel)	Yes
macOS (Apple Silicon)	Yes
Windows	No

There is no Windows binary. On Windows, run the CLI inside WSL.

Output for scripts

Every command prints its result formatted for reading. For scripts and agents, the read commands and the job creates take --output json and print the raw object instead; the agent skill uses that form.

Every command, flag and output shape: CLI reference.

The agent skill

The distil labs skill teaches your coding agent to run this workflow with you: choosing a task type, preparing data, running the pipeline, and analyzing what comes back. It follows the Agent Skills format, so one skill works in every agent that reads that format.

Install the skill

Install the skill:

distil skill install

The command needs git and nothing else. Run it again to update the skill. distil skill uninstall removes it.

It installs the skill for these agents:

Agent	Where the skill goes
Claude Code	~/.claude/skills/distil-cli
Codex, Cursor, GitHub Copilot, Gemini CLI, OpenCode, Cline, Zed, Warp, Amp, Antigravity, Droid, Kilo Code, Replit	~/.agents/skills/distil-cli, the directory these agents share

Other agents

Agents that only read their own directory, such as Windsurf, Goose and OpenHands, are installed with the skills CLI, which needs Node. Name the agent with --agent:

npx skills add distil-labs/distil-cli-skill --global --agent windsurf

Without --agent, the command finds the agents on your machine and asks which to install to. The skills README lists every agent name.

The skill needs the CLI and an account, the same as a run you drive by hand. To train your first model with your agent, start from the quickstart.

The REST API

Everything the CLI does, the REST API does. Use it when your work is already scripted in Python, or when you would rather call HTTP than shell out to commands.

The API is https://api.distillabs.ai. It takes a bearer token from the CLI, so it needs the CLI installed and signed in too. See Authenticate with the REST API.

• Using the REST API - the working guide, with the helpers every example depends on
• api.distillabs.ai/docs - the complete endpoint schema

The dashboard

app.distillabs.ai lists your models, datasets and traces. Each row opens a detail page with that entity’s status, its metrics, and the CLI command that works with it.

The dashboard is read-only: every job is started from the CLI, the agent skill or the REST API. It does not show credit balances; run distil credits-balance for those.

Next

Create your account, then build a model.