Agentastic Agent

Overview#

Agentastic is Agentastic.dev's own coding agent. It works like every other agent in the app: pick it in Agent Home, talk to it in a terminal or in Chat, ask it for a code review, or let a Manager Agent hand it work. The difference is that you don't have to install it first.

Getting it#

When no agentastic command is on your Mac, Agentastic.dev downloads the latest release in the background the first time it launches, so the agent is already in the picker when you first open it. The download comes from the same release feed the Homebrew formula is built from, arrives over HTTPS, and is checked against its published SHA-256 checksum before it is used.

An installation you already have always wins. To manage the CLI yourself, install it with Homebrew:

Agentastic.dev then uses your copy, in Agent Home and in terminal panes alike, and stops downloading its own.

In Settings > Agents, the Agentastic tile reads Included with its version while the app is running its own copy. To pin a particular installation, choose Configure… on the tile and pick it under Executable Source.

Staying up to date#

• The included copy checks for a newer release each time Agentastic.dev launches. When its tile shows an update, clicking it fetches the release in place; there is no installer to run.
• Your own copy updates the way you installed it, for example with brew upgrade agentastic.

Downloaded releases are kept in ~/Library/Application Support/Agentastic.dev/AgentRuntimes/.

Using it from Terminal#

When you have no installation of your own, terminal panes in Agentastic.dev put the included copy on PATH, so agentastic works in any pane.

To use it outside the app too, choose Agentastic > Command Line Integration… and click Install agentastic under Agentastic CLI. That writes a small launcher to ~/.local/bin/agentastic that runs the included copy even when the app is closed and follows its updates. It never replaces an agentastic already at that path. Put ~/.local/bin first on your PATH to use it; the sheet reminds you when it finishes.

Choosing a model#

The model chip in the Agent Home prompt and in the Chat composer opens Agentastic's model picker. It lists the models available through your providers and remembers them between launches, so it opens instantly and refreshes in the background; the status line reads refreshing… while it does.

• CLI Default names what Agentastic uses when you don't choose, for example "Agentastic currently uses Muse Spark · Medium thinking."
• Auto, pinned to the top of the model list, lets Agentastic choose among your available models. It has no thinking-effort setting of its own and appears once there is more than one model to choose between.
• Thinking Effort sets how hard the chosen model thinks, for models that support it.

What it supports#

Feature	Agentastic
Chat	Native structured Chat, including sub-agent conversations
Auto-approve	--dangerously-bypass-approvals-and-sandbox
Resume	resume --last
Initial prompt	Yes
Code review	agentastic review
Plan mode	Not available

See Supported Providers for how Agentastic compares with the other agents.