Agents in Settings

Settings > Agents is where Tori finds your agent CLIs and where you set each one up. Open Settings (the gear in the title bar) and pick Agents in the rail.

The agents list

The page opens on a table of every agent Tori knows, with a search box and a Check again button that looks for them again. Each row shows:

The State of an agent:

Agents are looked up on your login shell’s PATH, so one installed through nvm, volta, asdf, mise or Homebrew is found the same way your terminal finds it.

Turn an agent on or off

The switch at the end of a row, Offer Claude in Tori, decides whether the agent is offered anywhere: the model palette, the launch menu, the + for a new chat, the Autopilot’s picks. Turn off the ones you do not use so they stop cluttering those lists.

An agent can only be turned on once it can be used. The switch explains why not: Install it first, Sign in first, Add an account first, or Still being checked.

A chat already open on an agent you turn off keeps its tab, and its composer says the agent is off.

An agent’s page

Click a row to open the agent’s page. Agents at the top, or Escape, takes you back to the list.

The header shows the agent’s mark, name and version, the path it was found at (or not found on your login shell), its state, and the same on or off switch.

Set it up

An agent that is not ready yet shows a Setup group with three steps and how many are done:

The refresh button on the group checks again, for when you installed or signed in from your own terminal.

Install and update commands Tori knows:

npm’s global bin folder has to be on your login shell’s PATH for Tori to find what npm installed.

Update and uninstall

When the installed version is older than the one Tori was built against, the page shows Update available with the vendor’s update command and an Update button. A newer version than Tori knows says nothing; that is normal.

Further down, Update and Uninstall run the vendor’s own commands in a terminal tab. Uninstall really removes the binary.

An agent that does not report its version says It does not report a version, so Tori cannot check it.

Notes on the page

• Chat runs through npx, which is not on your login shell’s PATH. The terminal works without it. Codex and Pi chat through npx. Install Node.js to fix it.
• ANTHROPIC_API_KEY is set in this environment, so Claude bills against that API key rather than the subscription it is signed in with. Shown when the agent says it is using an API key from your environment.

Accounts

For an agent that has accounts, the Accounts group lists one card per login. See Accounts and usage for adding, renaming, signing out, the default account, and each account’s quota chips and warnings.

Models

The Models group lists every model the agent says it can run, with its id and its effort levels. With two accounts there is a tab per account, because each account can be offered different models. The heading shows the plan and the number of models.

• Search filters a long list by name, id or effort level.
• Ask again asks the agent for its models now. Use it after a new model is released.
• A model marked yours is one you configured in the agent’s own settings; Tori passes it through as is.

Its own options lists any other switches the agent publishes, such as a fast mode. They are set per chat, from the composer’s bar.

Files

For an agent whose files Tori knows (Claude today), the Files group lists what each account’s home folder holds, one tab per account:

Each row says what is there: on disk, not created, a file count for a folder, or broken link with where the link points. A linked row shows the link’s target beside its path.

Click a row to open it. Then:

• Open opens a file in the editor.
• New creates the file, or for a folder asks for a name and creates a new skill, command, subagent or rule from blank. Press Enter or Create, Escape to cancel.
• Write with an agent opens a new chat draft in the selected project with the file attached and a first line such as Write my Claude skills., for you to finish and send.
• Reveal in Finder shows it in Finder.
• Inside a folder, each item has its own open button and a remove button. Removing asks first, shows the full path, and cannot be undone. Removing a skill removes its whole folder, so no half skill is left for the agent to load.

The actions that open something in a project need a project selected in the sidebar first; until then they say Select a project first.

Tori lists the settings file but never edits it for you. Open it to change permissions or hooks by hand.

Plugins

For Claude, the Plugins group lists the plugins installed for each account, with their version and where they came from. A disabled plugin is marked off. Plugins are installed and removed with Claude’s own commands.

Chat capabilities

The Chat capabilities group says what this agent can do in a Tori chat, as chips: steering, rewind, approvals, exact diffs, spending ceilings, subagent lanes, attachments, history. Hover a chip for what it means.

What the agent cannot do is shown struck through, and listed under the chips in full sentences, for example why messages typed during a Codex turn wait for the next one. An agent with no chat at all says Terminal only. Tori has no chat transport for this agent.

See Supported agents for the same comparison across every agent.

Sessions

The Sessions group says where Tori reads this agent’s past sessions from, such as ~/.claude/projects, or that they come over the agent’s own protocol. Chips show live status when the agent reports its state through its own hooks, and needs-you when Tori can tell it is waiting on you.

Binary path

To use a particular copy of the agent, for example a wrapper script or a build you are testing, type its path in Binary path. New chat sessions use it instead of the one found on your PATH. Leave it empty to go back to the one found. When an adapter file of your own overrides the agent, the page says so under the field.