Choose where Lea runs

Lea is always reachable: her chat, her inbox and her settings live in Minga. Where she runs, the place each of her turns is computed, is your choice, per agent.

Choice	Where her turns run	Billed by	Good for
On a runner	The Claude Code or Codex CLI on one runner and profile you pick	Your CLI setup’s provider	Running her turns with your own CLI setup
Any of my runners with Claude Code or Codex	Any of your runners that has that CLI	Your CLI setup’s provider	Not depending on one machine
In the cloud	Minga’s cloud, with your workspace’s Anthropic or OpenAI API key and the model you pick	That provider, on your workspace’s key	Working with no runner online

Set it

1. Open Project settings › Lea (for another agent, Project settings › Agents and the agent).
2. Under Where Lea runs, pick one of the three choices, then the runner and profile, the CLI, or the provider and model.
3. Check the line under the picker. It says what follows: “Away while my-mac is offline”, “Thinks on any of your runners with Claude Code; resumes its session only on the same profile”, or “Thinks in Minga’s cloud with the workspace’s Anthropic key; no runner needed”.

You can also choose this when you create a project: under More options, Where Lea runs. If you leave it, she runs on your first runner, or in the cloud if you have no runner but a key. With neither, she waits for any of your runners with Claude Code, and her chat says so. An agent runs only on runners of the person who looks after it.

For the cloud: add a key

Settings › Model providers holds one Anthropic and one OpenAI key per workspace. Owners and admins add them. Minga tests a key by listing the provider’s models (no tokens are spent) before saving it, stores it encrypted, and never shows it again.

Her tools

Lea’s turns plan; they do not edit your repo. By default they load only Minga’s tools. On a runner you can turn on Use your CLI’s connected tools in its turns (in New project it reads “in Lea’s turns”), so she can call the MCP servers you set up in your CLI (an analytics or issue tracker server, say) while she plans. Off, she puts work with outside services into items for workers.

When Lea can’t work

There is no silent fallback between the choices. If the place she runs is not available, her turn waits and her chip tells you why:

Chip	Why	Fix
away	She runs on a runner, and no runner she can use is online	Start the runner (minga runner start), wake the machine, or pick another choice
queued	Her turn waits for something else, for example a runner that is busy or rate limited	Wait, or check Settings › Runners
can’t work	She runs in the cloud, and the workspace has no key for that provider, or the provider refused it	Add or fix the key in Settings › Model providers
can’t work	The person who looks after her left the workspace	Take her over in Project settings › Lea

Your messages are saved either way. She reads them as soon as she can run again.

Note

A runner must run a recent enough version for her turns. If the picker says a runner “runs 0.5.1: update to use it”, run minga update on that machine.