How gezel runs
Gezel is two cooperating pieces:
The app — the window you see. It shows your crew, projects and threads, and talks to the service over a private, authenticated local connection.
The service (
gezeld) — a background process on your machine that owns everything real: your files, your threads, the connections to AI providers. It keeps working when the window is closed, which is what makes scheduled jobs and long-running tasks possible.
The app finds or starts the service automatically. On a standard install the service runs machine-wide; there is also a per-user mode, and Settings lets you enable start-at-login so Gezel is always on duty.
How Gezel acts
A language model on its own can only produce text. Gezel gives it tools — reading and writing workspace files, saving memories, running scripts, creating tasks, messaging others on the crew. Tools are how a model's words become work, and every tool call is subject to the same consent and audit rules regardless of which AI provider is behind it.
The Tools and toolsets article lists every group; each role article shows exactly which groups that role carries.
Providers
The service speaks to AI providers through one pluggable layer: GitHub Copilot, OpenAI, Anthropic, and local engines all plug into the same seam. That's why switching providers — or giving one crew member a different model than the rest — is a setting, not a migration.