Skip to main content

​
Zero to Agent in 5 Minutes

This guide gets you from zero to a working agent in 5 minutes.

​
1. Send a Message

Type anything in the chat box. Wolffish assembles context from your workspace files, calls the LLM, and streams the response. Try:
If shell capabilities are enabled, the agent will call the shell_exec tool, and you’ll see an approval dialog for potentially dangerous commands.

​
2. Inspect What Happened

After the conversation, check these files:

​
3. Add a Capability

Create a new folder in brain/cerebellum/ with a SKILL.md file:
The SKILL.md frontmatter registers the capability with the system. The markdown body contains instructions the LLM reads at runtime. See Creating Capabilities for the full guide.

​
4. Edit Your Agent’s Personality

Use Wolffish’s built-in file viewer to open brain/identity/soul.md and edit it directly in the app. You can also use any external text editor. Changes take effect on the next message — no restart needed.

​
5. Configure Providers

In Settings → Models, add API keys for the providers you want. Then pick your Brain model (provider + model) from the model switch beside the chat input — that’s the one that runs. There’s no automatic fallback to other providers: the model you pick is the model that answers. To run offline, flip the switch to Local (Ollama). Want parallel multi-agent work? The same model card carries a Single / Workflow chip row above its search — pick Workflow and the agent can plan phases and spawn parallel agents on any model you’ve connected. The thinking effort dial lives in that panel too, so every model knob is in one place.