browse docs
on this page

Agents tab

The Agents tab builds reusable agent personas and prepares runs from them. A persona captures how you want an agent to work once, so you can launch it again without re-describing it every time.

Build a persona

A persona is a name, a provider, a system prompt, and the KödSkills it should lean on. Personas are kept in an app-wide or per-workspace list, so you can share a persona across every project or scope it to one.

The persona document is versioned and never overwritten when it can't be read, so a downgrade or a corrupt file can't wipe your agents.

Attach Connections

A persona can attach Connections — the MCP servers an agent should reach — from a curated catalog or a custom stdio or remote server you enter yourself. Ködade stays bring-your-own-key: a catalog entry only tells you which token or OAuth client to set up in your own CLI config, and there are no key fields anywhere.

Prepare a run

Preparing a run drafts a normal KödWork background task from the persona and hands off to the existing spawn path. A launched persona therefore keeps the same durable progress, scoped permissions, review, and recurrence as any other KödWork task.

When a persona relies on KödSkills, those skills install through the same reviewed KödHarness flow as run preparation — nothing is written silently. If an attached connection isn't installed for the chosen provider, preparing the run warns you without blocking.

What a persona can and can't do

An attached connection can't let a run do anything a KödWork task with the same CLI config couldn't. Personas describe intent and configuration; the agent CLI you install and authenticate still does the work, under its own permissions.