Skip to main content
Providers settings control the coding agents Traycer can run on a host. They apply to the host selected with the host switcher at the top of the Settings sidebar; see Host Picker. For which coding agents Traycer supports and what each one can do, see Agents & Models.

Provider List

The list on the left shows every provider, with the enabled ones first.
  • Search providers narrows the list by name.
  • Filter providers shows All, Enabled, or Disabled providers.
  • Refresh all providers checks sign-in and CLI status again.
Select a provider to open its tabs.

Turn A Provider On Or Off

The switch in the provider’s header reads Enabled or Disabled. Only enabled providers appear in the composer’s model picker.
  • Claude Code and Codex are enabled in a new install.
  • Other providers, except Traycer Inference, can turn on automatically when the host starts and verifies their credentials. If verification fails, a later host start can try again. Signing in during a session does not turn a provider on by itself.
  • Once a provider has an enabled or disabled setting, host startup preserves it. You can change it with the switch.
  • Traycer Inference is opt-in. It uses Traycer credits, so only you can turn it on.
At least one provider must stay enabled. A provider with managed profiles needs at least one enabled profile before you can turn it on.

Reset Unreadable Provider Settings

If Traycer cannot read a provider’s saved settings, a warning in Providers offers Reset <provider> settings. Use it when the warning prevents you from saving that provider’s key, on/off switch, or CLI choice. Read the confirmation before resetting. Traycer keeps a copy of the settings file in the same folder, then resets only that provider. Its saved key, CLI choice, environment overrides, and launch arguments stop being used; other providers’ settings are kept. After the reset, turn the provider on if needed and enter its key again if you use a stored key. Reapply any custom CLI choice, environment overrides, or launch arguments you still need. The result message names the backup. If it asks you to restart the host, restart the host selected in Settings to finish applying the reset.

Provider Tabs

Tabs depend on the provider. A provider opens on its first tab. Usage limits and the banners that appear when a limit is reached are covered in Usage limits and model routing.

Claude Code Session Names

In Providers → Claude Code → CLI & Args, the Session name group controls the names of new Claude Code sessions on the Terminal interface. It applies to the host selected in Settings. Edit Name template, check the sample preview, then press Enter or leave the field to save. Reset restores Traycer · {project} · {date}. For example, {project} · {branch} · {date} includes the working folder, branch, and date. Missing values are left out; new Tasks and agents may not have titles yet. Resuming a session keeps its existing name, including a name set with Claude Code’s /rename. Changing the template does not rename existing sessions or Chat interface agents.

Account

The Account tab holds the API key for providers that authenticate with one. Paste a key and click Save. A stored key is encrypted on the host and overrides the one in your shell. When a key is stored, the field reads Replace stored key…, and Clear removes it. Some providers show a Create an API key link to their dashboard.

Managed Profiles

Claude Code, Codex, Grok, and Antigravity can hold more than one account. The provider’s own CLI login, the one you would use in a terminal, is the Terminal account. Managed profiles are extra accounts next to it.
1

Add the profile

On Profiles & Limits, click Add profile.
2

Name it

Give the profile a name and a color.
3

Link the account

Click Link account and sign in to the account this profile should use.
Each profile has its own switch, so you can turn one off without removing it. A profile’s menu lets you rename it, change its color, sign in again, or remove it. What a new profile shares with the Terminal account depends on the provider: Grok profiles use separate GROK_HOME directories, so each profile keeps its OAuth account, configuration, and sessions isolated. New profiles start with an empty private store: ambient configuration, MCP credentials, plugins, skills, memory, and sessions are not copied. Ambient xAI API keys and external-auth commands are not inherited by managed Grok profiles. Antigravity’s Terminal account can use Google sign-in or inherit GEMINI_API_KEY from your shell. A managed profile uses its own Google sign-in or stored Gemini API key; it does not inherit the shell’s key. Signing a profile out never touches the Terminal account. To pick a profile for the next turn, use the profile dropdown in the composer’s model picker. It appears once a provider has two or more profiles.

Signing In Again

When a provider is signed out, the composer shows a banner above the input.
  • Authenticate (for example Authenticate Claude Code) opens your browser to sign in. If the browser doesn’t return on its own and shows a code, paste the code into the banner.
  • Providers that sign in from a terminal offer a terminal sign-in instead.
  • “Or paste an API key or token” takes a key or token directly, for providers that accept one.
  • Check sign-in status checks again after you sign in somewhere else.
If the agent’s profile is signed out, the banner offers Continue on Terminal account and Sign in. If the profile no longer exists, it offers Manage in Settings. Browser sign-in runs on the machine the host is on. For a host on another machine, sign in with that provider’s CLI there.

CLI Binary And Arguments

The CLI & Args tab chooses which binary Traycer runs for the provider:
  • the copy Traycer manages for you,
  • a copy found on your PATH, or
  • a binary you point to with Add custom path.
For CLIs that Traycer manages, open the version menu on the managed row to see the available versions.
  • Download, Use, and Delete act on one version.
  • Use pins that version. A pinned provider shows “Pinned to” its version; Use latest automatically removes the pin.
  • Check for updates looks for a newer version, and Auto-download updates downloads new versions as they come out.
Traycer does not manage the Amp, Hermes Agent, or Command Code CLIs; install them yourself, then pick the detected copy or add its path. For Claude Code, Codex, and OpenCode, this tab also has Terminal interface CLI arguments. They are appended to the CLI when you start an agent on the Terminal interface, and they pre-fill the launch picker, where you can change them for one launch. Use bypass or auto-approval flags deliberately. Codex permission flags such as --approve-for-me, --ask-for-approval, and --sandbox apply when a session starts; when Traycer reopens or forks a Codex session, the session keeps its own permissions, so Traycer leaves those flags out.

OpenCode Versions

In Providers → OpenCode → CLI & Args, choose Managed 2.x to use OpenCode 2, or Managed 1.x to use OpenCode 1. The same OpenCode binary choice is shared by Traycer Inference, OpenRouter, and Hugging Face. Your choice also applies to existing chats from their next message, not to a turn already running. A chat keeps its current version while background work is running, the selected binary is not ready, or the other version cannot run that message with its selected model, attachments, or commands. The composer explains why it has stayed on its current version.

Native Subagents

On CLI & Args, Allow native subagents is on by default. Turn it off to direct this provider’s agents to delegate through Traycer agents instead of the provider’s built-in subagents. Each provider has its own setting on the host selected in Settings. If the switch is missing, update that host to a version that supports it. Traycer disables the built-in subagent tools where the provider supports it. For other providers, the agent receives instructions to use Traycer agents instead; this is not a tool-level restriction. Chat agents pick up the change on their next turn. Restart a running Terminal interface agent for it to pick up the change.

Model List Timeout

The Model list group on the CLI & Args tab sets how long Traycer waits for the provider to list its models: when you open the model picker, and when an agent creates another agent. Timeout offers 60 s (the default), 90 s, 120 s, and 180 s. Pick a longer one on a slow or heavily loaded machine, where a provider can take more than a minute to answer. With Same for all providers on, the timeout is shared: changing it changes every provider whose switch is also on. Turn the switch off to give this provider its own timeout, which starts at the shared value; turn it back on to follow the shared value again. A change applies from the next model list the host reads, and a model list that arrives after the wait is still kept for the next request. The group is hidden on a host running a version without per-provider timeouts.

Environment Variables

The Env tab sets environment variables for one provider. Each row has a name and either Set with a value, or Unset to drop a variable inherited from your shell. They apply when Traycer starts that provider. For Cursor, these variables reach only its configuration commands, such as MCP setup, and not chat turns. Shell-wide host environment variables live in Shell.

Permissions

The Permissions tab shows who reviews this provider’s commands in Auto mode. Where the provider has a reviewer of its own, a switch chooses between it and Traycer’s judge; otherwise the tab shows which one applies. All permission settings opens Permissions.

Model Providers

OpenCode reaches models through upstream providers, which you manage on its Model Providers tab.
  • Connect a provider by signing in or with an API key. The dialog lists the sign-in methods that provider offers.
  • Disconnect removes a provider’s credential.
  • Add custom provider adds an OpenAI-compatible endpoint: a Provider ID, Display name, Base URL, optional API key, its Models, and optional Headers.
  • Manage Go opens your OpenCode Go account.

MCP Servers

Add and check provider-specific tool servers in the MCP tab. See MCP servers.

Plugins

Click Add from source, enter a Source (an npm package, a path, a git URL, or plugin@marketplace), and click Install. The location picker at the top chooses whether a plugin applies to every workspace on the host or to one workspace. Restart the provider’s running agents for plugin changes to reach active sessions.

Skills

Import or write reusable instructions in the Skills tab. See Skills.