{AF} AgentForge OpenClaw setup, guides, and ready agents
Install guide Book setup

First agent

How to build your first useful agent in OpenClaw without starting from an idea that is too big

The first agent does not need to be impressive. It needs to be useful. The best path is starting from a stable base, choosing a sensible model, and validating one concrete task before adding channels, plugins, or automation.

Step 1: healthy base and a working dashboard

openclaw onboard --install-daemon
openclaw doctor
openclaw dashboard

If this does not work, the problem is not the agent but the setup. The first agent must be validated on top of a healthy base.

Step 2: choose a model that fits the job

openclaw models list
openclaw models status

The OpenClaw docs recommend starting with the best model available for the task and using fallbacks only when cost or latency require it. If you are using Ollama, first confirm the model is actually visible and useful for tool workflows.

Step 3: validate one single task

  • do not start from a “universal agent”
  • pick one repetitive task that is easy to verify
  • measure whether the result really saves time

Good first tests include email support, web-research note collection, simple follow-up, or recurring reporting.

Step 4: only then add plugins or channels

Many users make the flow too complex too early. First make one agent work inside the dashboard or Control UI. Only then is it worth adding plugins, channels, or extra integrations that increase the surface area for failure.