Brains and agents

The brain is who answers a chat. Pick it in the composer’s brain picker (tooltip “Choose who answers”) before the first message.

The three kinds of brain

Brain	Billed to	Where it runs	Best for
Optimal	Wallet credits	MIOSA, working on your machines through scoped tools	Questions about your workspace, building, investigating
A harness	Your own model account	On the sandbox or computer you pick	Coding with Claude Code, Codex, OSA, and others
A saved agent	Your own model account	On the sandbox or computer you pick	An agent with your own instructions, tools, and permissions

Optimal

Optimal is the model built into MIOSA. The picker shows it as “Built into MIOSA” and “Uses your wallet credits, about 4 credits per message”.

It is one model, with a 1M-token context.
You cannot pick a different model for it, and it does not use your own model keys.
It is the only brain that offers Investigate mode.

On the free plan with no credits left, Optimal is disabled in the picker with “Add credits to use Optimal”.

Harnesses

A harness is a coding agent that runs on a machine on your own account. MIOSA credits never pay for its model. The picker lists the harnesses under Your agents:

Claude Code, Codex, and OSA;
OpenCode, pi, Prime Agent, Kimi Code, Mistral Vibe, and Hermes, when your organization can connect them.

Each row shows its connection state: “Connected” with the account, “Not connected”, or “Needs signing in again”. Connect or Reconnect opens the sign-in for that harness, then returns you to Optimal. See Harnesses and auth for what each harness accepts.

A harness always asks where to work, in any mode, because each message is a run on a machine. The cost strip says the work is billed to your own plan.

Saved agents

A saved agent is an agent you built in Agent Studio, with its own instructions, harness, model, tools, and permissions. Saved agents appear under Your agents next to the harness they use. An archived agent is not listed.

To build one, choose Build an agent at the bottom of the brain picker, or open Agent Studio. From the CLI:

miosa agent new reviewer --harness claude-code --instructions-file review.md
miosa agent list

Switching brains

The brain is fixed after the first message. The picker’s tooltip then reads “This chat keeps its brain. Start a new chat to change it.” Picking another brain in a started chat starts a new chat.

Chat with an agent on a machine: the same brains from a computer or sandbox page.
Agent confinement: what an agent can reach from inside its machine.
Pricing and usage: how each brain is billed.