Say it. Your Mac does the work.

Solo is free with every capability for one profile. Bring a provider key, a local model, or both; a provider's own bill stays yours.

Paid plans add profiles to an account. Bandicot does not resell model tokens.

Bandicot gives every agent a project, tools, a session that survives restarts, approvals, and a clear place to stop.

The native host owns execution facts. The model proposes work; it does not certify a write, spend, or success.

Keep files, context, plans, and decisions with the project instead of losing them in a disposable chat.

A session is anchored to a project boundary so a pivot starts from an intentional context, not accidental history.

Close the window, restart, or pick up from another of your Macs without turning the work into a brand-new conversation.

The host records session state and every window reattaches to it. When a long chat is summarized, the agent can still search and reread what was there.

When work can change files or systems, Bandicot makes the decision visible before the effect lands.

The approval boundary fails closed when an action is not allowed. The model cannot grant itself permission.

Hold the mic or press a key and say what you need. Say stop to end a run, wait to hold it, yes to approve.

Speech becomes words on your Mac, offline, in English or Spanish. Someone else talking nearby is shown struck through and never sent.

Once a local model is downloaded, chatting with it needs no connection.

Small local models run through llama.cpp on your Mac. Hosted providers still need the network.

Let an agent inspect, plan, and use tools inside the project you selected instead of an unbounded chat transcript.

The host controls tool calls, cancellation, permissions, and the final recorded result.

Every change an agent makes is a version: undo a whole run, restore one file, or drag a slider to put a folder back to any moment.

Works with or without git, on any disk you plug in. Restoring is itself recorded, so an undo can be undone.

Use a local model, an OpenAI-compatible endpoint, or a hosted provider you sign in to, and switch per session without rebuilding the project.

Provider adapters keep model choice separate from project state. Remote providers receive the prompts sent to them.