Installation & Setup
If you don’t have an account yet, you can register for one here.
Install CLI
Download the CLI and authenticate with Ellf.
Connect
Connect Ellf to your coding assistant like Claude Code.
Set up your cluster
Launch your private data processing cluster.
Installation
Make sure you have Python 3.10+ installed and use the following command to install the client. It’s recommended to use a virtual environment.
$pip installellf-cli
This installs the CLI that you and your coding agent can use to set up projects with Ellf, run tasks, actions, agents and services, and manage your cluster. Next, you can authenticate and connect Ellf to your coding assistant.
Coding assistant integration
Ellf currently supports Claude Code, Codex, GitHub Copilot and OpenCode out of the box. You can install integrations for one or more tools as comma-separated names in the --agent argument on login. This will download and add the skills and MCP server.
To update your skills, simply re-run the login command with the --agent argument. Also re-run the command whenever you’re creating a new virtual environment.
Notes on differences
The supported coding assistants slightly differ in how they handle skills, commands, MCP servers, debug logs and asking users follow-up questions. In Codex, commands are available prefixed by $ instead of /, e.g. $ellf-project. OpenCode v1.x doesn’t support commands but can still trigger skills if their name is used.
Other coding assistants
In order to use Ellf, your coding assistant needs to support adding custom skills via Markdown files and MCP server integration. If your coding assistant of choice is compatible, you can always copy over the skill files manually and add the Ellf MCP server.
https://app.ellf.ai/ellf-core/mcpNote that we’ve currently only tested Ellf with the coding agents and models listed here, so we cannot guarantee that it will work as intended with other models and setups.
Configuration
To edit your organization settings, select Settings in the app sidebar. This is also where you can manage and invite users. Alternatively, you can also use the users CLI. Under Account in the top bar’s user dropdown menu, you can configure your account details and personal preferences.
Configuring Prodigy for annotation
The Prodigy annotation tool included with Ellf supports various global and interface-specific configuration settings in JSON format, to customize the app’s style and behavior. In Ellf, they can be defined per project or per task, in the web app or CLI:
| Project | ProjectSettings | --prodigy-config in projects create/update |
| Task | Annotation settings | --settings.config in tasks create (if defined) |
Project-level overrides are the default for all projects and can be overwritten by the task config. Note that Ellf only supports overwriting config settings for personal preferences like user experience and interface style, not config settings that modify the annotation task and logic itself.