Agents and models

Supported agents

OpenOrc supports Codex, Claude Code, OpenCode, Oh My Pi, Cursor, Pi, Antigravity, and Vercel Labs fx. Install the command-line agent you want, connect its account or API key, then enable it in Settings → Connections. OpenOrc uses that agent's own models, configuration, and session storage.

Each name below links to the agent's installation guide. Model charges and account allowances belong to the provider you connect.

How they differ

Every agent supports conversations, streaming responses, tools, attachments, cancellation, saved sessions, and OpenOrc's project tools. The picker and action menus offer the capabilities that the installed agent can use.

All agents can be controlled from a paired phone while the connected computer is available. Cloud support means the agent can also run in an OpenOrc cloud workspace. Oh My Pi uses provider API keys in the cloud.

Cursor queues messages until its current turn ends. fx accepts text steering; images wait for a new turn. Other messages may wait while an agent is asking a question, awaiting approval, or compacting.

Codex, Claude Code, OpenCode, Oh My Pi, and Pi support native manual compaction. With Cursor, Antigravity, and fx, Compact context starts a fresh session with a short handover brief. It preserves less context than native compaction.

Terminal session import is available for Codex and Claude Code. Those two also support background memory summaries; other agents can use shared memory directly, or you can choose a supported summary provider in Settings.

Modes and permissions explains each agent's policy. Claude Code cannot use Plan or Accept edits on Windows; fx does not expose a read-only Plan mode.

How OpenOrc finds them

On macOS and Linux, OpenOrc asks your login shell where each agent is, the way a terminal would, so it finds the same codex, claude, opencode, omp, cursor-agent, pi, agy, and fx you use there. Then it checks the installed version and the agent's sign-in or available models, without asking a model to work. It checks when it starts and again whenever you ask.

1. Install the agent or sign in from a terminal.
2. Open Settings → Connections and click Refresh connections. During setup, click Rescan instead.
3. Check the agent's status. Connected means it's ready. Not logged in shows the sign-in command to run, and Not installed links to the install guide.

If Codex or Claude Code signs out while you work, the thread shows Sign in, which runs the agent's sign-in command in the thread's terminal.

Choose the agent and model

Every thread has its own agent and model. To change them, click the model name in the composer's bottom row.

• The rail on the left switches between your enabled agents. On iPhone, use the icon tabs; the selected agent's name appears above them. An agent that isn't installed or signed in is marked, and its models stay unavailable until you fix that.
• Search models filters the list. Older models wait under Legacy models.
• A model the installed CLI can't run stays in the list with the reason, such as a model that needs a newer Claude Code.
• OpenOrc asks each agent for its models and keeps the list for 5 minutes. Refresh models asks again.

Only agents enabled in Connections appear in the model picker and can start work. Turn one off to hide it, or open Add agents to enable it again. New threads start with the agent you picked during setup, or the first agent that's signed in. The New thread page then remembers your last choice for each project.

Effort and Fast mode

Under the model list, Reasoning effort sets how much the model reasons before it acts. Each model offers its own levels, such as Low, Medium, High, and Extra high. Higher effort usually takes longer and uses more of your allowance. Choosing another model resets effort to that model's default, and Reset to model default does the same for the current one.

With Claude Code 2.1.278 or newer, models that offer Extra high also offer Ultracode. It runs at Extra high and tells the model to orchestrate dynamic workflows by default.

The lightning button in the picker's rail requests Fast mode, which trades a higher cost for faster output. It appears only for models and accounts that can run Fast. A lightning mark beside the model name shows it's on. If Fast mode is on but can no longer run, the picker says why, such as This Claude plan does not include Fast mode.

Switch agents in a thread

You can give a thread to another agent at any time: pick one of its models in the picker, and your next message goes to it. Agents can't read each other's sessions, so the new agent starts a fresh session with a short brief: your last four requests and the previous agent's last reply. The composer tells you so before you send.

• In a project thread, switching back to an agent that worked there before resumes its own earlier session, without a brief.
• In the Workspace, every change of agent or folder starts a fresh session with a brief.
• While the agent is working, a message for the new agent waits in the queue until the turn ends.

Changing the model, effort, or Fast mode within the same agent applies to your next message without starting over.

Use several Claude accounts

OpenOrc can run Claude Code on more than one Claude account, such as a work account and a personal one. Your existing Claude Code login is the first account. Each account you add keeps its own Claude Code login, settings, sessions, and usage. It shares the personal skills, plugins, and memories of your existing login.

1. Open Settings → Connections and find Claude accounts.
2. Type a name, such as Work, and click Add account and sign in.
3. Sign in to that Claude account in the browser window that opens. On macOS and Linux, when Chrome is installed, the sign-in opens in a Chrome Incognito window, so your browser's own Claude login stays as it is.

Each account's row has Make default, Sign in again, and Disable. The default account is the one new threads start with. Changing it doesn't move existing threads.

• To start a thread on another account, pick it at the end of the line under the New thread composer. The choice appears for Claude models.
• To move a thread to another account, click the account's name in the thread's header and choose one. The switch waits until the current turn ends. Your next message continues the same conversation on the new account.
• When a turn stops on a usage limit, the thread offers your other signed-in accounts. Choosing one sends Continue where you left off. from that account.
• Schedules and Orclings each keep the account you choose for them.

Usage and allowances

The usage ring at the foot of the sidebar's rail shows how much of this week's allowance is left on the account your next message would use: the open thread's account, or the one the New thread page has chosen. Click it for a table of every account's 5-hour and weekly limits. Settings → Usage shows each account's limits in full, as the agents report them through their own CLIs.

• For Codex and Claude Code, each allowance window, such as 5-hour window or 7-day window, shows how much is used or remaining and when it resets. Open usage opens the account's own page.
• Codex also shows any usage resets you have. When Codex allows it, Use reset spends one after you confirm.
• Antigravity shows the model quota pools and reset times reported by its Google sign-in. API-key usage belongs to the Google AI Studio account instead.
• fx shows AI Gateway balances and budgets, plus Codex or Grok plan allowances and reported resets. The sidebar follows the thread's selected connection; Settings uses fx's saved connection. Custom connections use their provider's dashboard.
• Cursor allowances are on the Cursor dashboard. OpenCode, Oh My Pi, and Pi bill their configured providers; use those dashboards or the agent's native usage commands.

Settings also has Cost and Tokens views for OpenOrc's recorded history, where the agent reports those numbers. Missing usage or unknown prices stay unavailable. The limits include activity outside OpenOrc. When a thread hits one, its error links to Manage usage.

Update the agents

Settings → Connections lists each agent with its version and how it was installed.

1. Click Check for updates.
2. Click Update all, or the update button on one agent's row.

OpenOrc can update an agent installed with npm, with a Homebrew cask, or with the agent's own installer. For any other installation, and on Windows, it links to instructions instead. An agent that's installed but fails to start offers Reinstall. Agents keep working during an update, and each conversation moves to the new version on its next message.

Check for agent updates automatically looks every six hours while OpenOrc is open, using public version information from npm, Homebrew, or the agent's official release service. When it finds one, a notice offers Review updates.

Next steps