Skip to content

Providers and models

The app works with local coding-agent tools. It does not replace provider authentication or proxy provider traffic; it launches a local command or adapter and keeps workspace, tab, and review state around that session.

Chat UI means the app can render the provider’s messages, markdown, tool activity, and supported inline diffs. Terminal means the provider runs through its native CLI interface. Experimental integrations may change more quickly and may be hidden until enabled.

ProviderDefault commandChat UITerminalStatus
Claude CodeclaudeYesYesStable
CodexcodexYesYesStable
Cursorcursor-agentYesYesStable
OpenCodeopencodeYesYesStable
PipiYesYesStable
Oh My PiompYesYesStable
GrokgrokYesYesStable
Kimi CodekimiYesYesExperimental
GitHub CopilotcopilotNoYesExperimental
Factory DroiddroidNoYesExperimental
Kirokiro-cliNoYesExperimental
Qwen CodeqwenNoYesExperimental
HermeshermesNoYesExperimental
AntigravityagyNoYesExperimental
Gemini LegacygeminiNoYesExperimental legacy integration

The installed app is the authority if this table and its provider picker ever differ. Provider capabilities evolve independently, and a provider can support a feature in Terminal before its Chat UI adapter does.

Install the provider through its official instructions, authenticate it, and run its command in a normal terminal. For example:

Terminal window
claude --help
codex --help
opencode --help
pi --help

The app discovers commands through your macOS login-shell PATH. Restart the app after changing shell startup files. If the command fails or requests login outside the app, resolve that with the provider before starting a session.

Open Settings → Agents to:

  • Enable or disable providers.
  • Check installation status.
  • Override an executable command or add launch flags.
  • Control which models appear in pickers.
  • Choose default model and reasoning effort where supported.
  • Select Chat UI or Terminal as the default surface for new agent tabs.

First-run onboarding emphasizes installed stable providers. Use More or Settings to reveal additional and experimental options.

Session defaults cascade global → workspace → project. A project can inherit the workspace provider while overriding only its model or profile.

Git and review actions have two routing groups:

  • Fast for bounded work such as inline commit, Commit & push, Commit, and Create PR/MR.
  • Thorough for review, conflicts, failed CI, blocked merges, comments, and requested changes.

Advanced routing can pin a provider and model per action. It can also give a normal click and -click different reuse, submission, or background behavior. See Git actions and pull requests.

Model lists and reasoning controls are provider-specific. The app shows only configured or visible models and applies a reasoning level only when the selected provider supports it.

Changing a model or reasoning level affects a live Chat UI session only when that adapter supports mid-session changes. Otherwise, the new selection applies to the next session. Check the provider header before assuming a running process changed models.

The optional naming model handles one-time branch and tab naming. It can differ from the model used for implementation work.

Choose Chat UI when you want rendered markdown, tool activity, inline diffs, attachments, app-level history, and integrated notification handling. Choose Terminal when you need the provider’s native interactive UI, a flag that the Chat UI does not expose, or a terminal-only provider.

Using both is normal: keep an agent chat beside a terminal that runs tests or development servers. The selected surface is a default; each new tab can choose another supported provider or tool.

When Compliant Claude view is disabled, local Chat UI uses Claude’s noninteractive CLI path. Confirm the active CLI account and billing behavior before long-running work.

Provider permission modes decide whether an agent asks before commands, file changes, or risky actions. The app exposes the modes a provider supports and warns before explicit bypass settings.

Use the least permissive mode that still fits the task. A background or auto-submitted action does not remove the provider’s approval requirements; requests appear in the relevant session and notification queue.

Profiles let one provider use separate accounts, environments, or launch flags. Claude and Codex profiles point to distinct CLI config directories:

  • Claude launches the profile with CLAUDE_CONFIG_DIR.
  • Codex launches the profile with CODEX_HOME.

Each profile has a name, config directory, optional launch flags, and color. Other providers may expose profiles for organization even when they do not use a dedicated config-directory environment variable.

  1. Open Settings → Profiles and add a profile under the provider.
  2. Select the provider config directory.
  3. If validation fails, launch the provider once against that directory, authenticate, and check again.
  4. Add optional launch flags and a distinguishing color.
  5. Choose the profile at global, workspace, or project scope.

The provider header and tab indicators identify the resolved profile so work and personal accounts are less likely to be confused.

If a provider is missing or unavailable:

  1. Run its configured command in a normal login shell.
  2. Confirm authentication through the provider’s own CLI.
  3. Restart the app after changing PATH or shell configuration.
  4. Check that the provider and intended models are enabled in Settings.
  5. Confirm the selected surface is supported in the table above.
  6. For Chat UI discovery, run sc chat providers from an app-launched terminal.

A provider that works in Terminal but does not appear in Chat UI usually lacks a Chat UI adapter or has that integration disabled.