Skip to content

← All guides

ZeroClaw Setup Checklist: Workspace, Model and First Task

ZeroClaw.net · Updated

Set up ZeroClaw in layers: install the runtime, connect a model, then add the tools and channel your task needs. This checklist helps you make those choices without repeating the installation commands.

If you still need to decide whether to switch, start with the OpenClaw replacement checklist. For the commands and download walkthrough, go to install ZeroClaw.

1. Choose your installation route#

  • Prebuilt release: use the archive for your operating system and CPU. You do not need Rust to run the packaged executable.
  • Source build: choose this when you need to change code or enable build features. Review the Rust and VPS preparation guide.
  • Container: follow the Docker guide, including its persistent storage steps.

Keep the executable and dashboard assets from the same release. Confirm the executable starts before adding model credentials.

2. Choose a workspace and memory policy#

Create a dedicated folder for the assistant's work. Start with disposable files, and keep personal documents and credentials outside it.

Decide what should survive a restart:

  • Files created for the task.
  • Conversation or memory data you want to retain.
  • Configuration and provider credentials.

Use the configuration guide to connect the workspace, agent and risk profile. Use the backup guide before changing an established setup.

3. Connect one model first#

Choose a hosted provider if you want the model to run elsewhere. Choose a local model if your machine has the memory and compute it needs. The runtime's memory use does not tell you whether a model will fit.

Follow the provider chooser, then send a short message before connecting Telegram. A model reply confirms that the provider, model identifier and credential work together.

ZeroClaw dashboard showing a real Gemini response

The captured conversation shows the assistant replying “Ready.” Use a similarly small request for your first connection check. Open the full-size screenshot.

4. Add one channel#

For Telegram, follow the bot and account authorization guide. Keep the bot token private and authorize your own account before inviting other users.

Check these separately:

  • The channel connects.
  • Messages reach the intended agent.
  • An authorized message gets a model reply.

Keep the dashboard local during setup. Add remote access only when you have a reason to use it and a plan for authentication.

5. Try a small workspace task#

In the dedicated workspace, create a disposable file named practice.txt. Ask the agent to read it first. If your chosen policy permits file changes, ask it to write a short line to that file, then inspect the file yourself.

This preserves a useful check from the earlier setup walkthrough: verify the actual file, not only the assistant's description of what happened. If the agent refuses or a tool is denied, inspect the configured permission rather than telling it to ignore the boundary.

On Windows, check the exact executable or tool your task needs. Avoid copying a broad command list that includes deletion commands or an unrestricted shell. Follow the security guide when choosing permissions.

6. Finish the setup#

Your first setup is ready for a longer trial when:

  • The intended config loads.
  • A direct prompt returns a reply.
  • Your selected channel reaches the right agent.
  • Workspace permissions match the task.
  • You know which data to back up and how to stop the process.

For a failed step, use the troubleshooting hub. Keep installation commands in the installation guide so you follow one consistent route.