Orca IDE vs Caprock: a cockpit for coding agents, and the ledger of what they did

Orca IDE is a cockpit: a free, MIT-licensed desktop app for macOS, Windows and Linux that starts Claude Code, Codex and some thirty other command-line agents, each in its own git worktree, with a code editor, a built-in browser, managed SSH servers and phone apps. Caprock is the ledger: it records every agent session on the machine whether or not it started it, prices each one at list price, and joins it to the pull request and its checks. Run your agents in Orca, in Caprock's tabs or in your own terminal; the ledger is the same.

What Orca IDE is

Orca calls itself an agent development environment. It is an Electron app from Stably AI, MIT-licensed, built around one git worktree per task: you fan a prompt out to several agents, compare what they did and merge the one you want. It runs any CLI agent and ships some thirty preconfigured, with VS Code's editor, a Chromium browser with a design mode, GitHub, Linear and Jira boards, a managed Orca server on each SSH host (its docs mark remote servers beta), and phone apps for iOS and Android. On 27 October 2026 it had 107,887 GitHub stars, no pricing page, and an enterprise page that is a contact form.

Orca vs Caprock, side by side

What each one sees: the agents it started, or every session

Orca shows live status for the agents it started: its docs say to start one "through the agent combobox rather than typing the binary by hand". A Claude Code you start in iTerm is not in its sidebar. Its session history scans each agent's transcripts afterwards, so "if the CLI writes a transcript, it shows up in the panel after the next scan", resumable, with no cost on that page (its docs, read 27 October 2026). Its usage tracking reads the same files on the local host; a user who runs agents on SSH hosts wrote that "the token/cost analytics are local-only file scrapers" (#12005, open).

Caprock reads the machine, not its own tabs. A Claude Code session started in your own terminal appears live, with what it costs; Codex, OpenCode, DeepSeek Harness and Hermes Agent come in from the history they write, sessions from before you installed Caprock included. Sessions that run in a vendor's cloud or inside CI are not observed; Caprock sees the machines it is installed on.

The phone: native apps or the browser

Orca has an iOS app on the App Store and an Android APK; its docs mark the mobile companion beta. A phone pairs through Orca Relay, which needs a sign-in, or directly over your LAN or Tailscale; its troubleshooting asks for the same Orca account on the desktop and the phone.

Caprock has no phone app. The phone opens Caprock in its browser, paired by a QR code, with no account. It reaches your computer from anywhere through a relay that carries only end-to-end ciphertext, ours or one you run. Unless you untick It can control sessions when you connect it, the phone starts agents, answers permission prompts, and commits and pushes.

Cost, plan limits and telemetry

• Orca tracks usage against the plan for 11 providers, with reset times and an 80% warning, and estimates cost from "a local price table, not a live bill", per account on the local host; its docs say to "Prefer the provider console for authoritative spend". Cost per worktree or per pull request is not in them. Users with several accounts ask for failover: "work still stops dead when the active account hits its 5-hour or weekly window" (#20512, open, read 27 October 2026).
• Caprock prices every turn of Claude Code, Codex, OpenCode, Gemini CLI and DeepSeek Harness per repository, and the Pipeline shows each pull request with its checks and what the agents spent on it. A session on Bedrock or Vertex is priced the same as one on the API (internal/cost/cost.go). Claude's and Codex's 5-hour and 7-day windows sit beside the cost, one per sign-in, with a warning before the limit; the claude you start in your own terminal is counted too. Every dollar Caprock shows is an estimate at API list price, as Claude Code's own /usage is: not a bill, and on a flat plan not money saved.
• Telemetry: Orca sends anonymous usage events, such as which kind of agent you start (never prompts or output), to PostHog in the US; a switch in Settings, DO_NOT_TRACK=1 or ORCA_TELEMETRY_DISABLED=1 turns it off. Caprock has none.

Permission prompts by default

Orca pre-fills each agent's permission-bypass flag for new launches, such as --dangerously-skip-permissions for Claude Code, and its docs warn that a worktree is not a sandbox. Caprock starts new sessions accepting edits and asking before a command; bypass is one pick away in the New agent sheet.

Worktrees, tries and review

Much of the daily loop is the same in both, read from each project's docs on 11 October 2026:

• Ready worktrees. Both copy the ignored files a repository lists in .worktreeinclude, such as .env, and run a setup command such as npm install in a new worktree.
• Review back to the agent. Both let you comment on diff lines and send the notes to the agent, and both hand a refused commit, conflicts or a pull request's failing checks to an agent with one button.
• Tickets. Orca links GitHub pull requests, Linear and Jira issues and GitLab merge requests to a new worktree; Caprock starts a new copy from a GitHub issue or pull request.
• Several tries at one task. Orca's recipe races different agents on one prompt, a worktree each, and you review each diff. Caprock gives one agent the same first message in two to four copies and sets them side by side: files and lines changed, what each has cost so far and how long it has run. Orca's docs do not show cost per worktree.
• Clean-up and sleep. Both list worktrees with their size and git state for removal, and both can keep the computer awake while agents work.

Download and install size

Measured on 11 October 2026 from the files each project publishes on GitHub: Caprock v0.100.0 and Orca v1.4.224. Orca ships its own Chromium; Caprock uses the web view the system already has, except in the Linux AppImage, which carries its own WebKit.

Installed on one Mac, Caprock.app takes 47 MB and Orca.app (1.4.220) 624 MB, by du. These are sizes on disk. Neither app's memory use or speed has been measured against the other.

Orca vs Conductor, cmux, Herdr and T3 Code

• Conductor is a Mac cockpit for the same job: free for local workspaces; Pro, $50 a person a month, adds cloud workspaces and the iOS app; Teams, $60 a user a month, an admin portal and central billing; cost per person or per repository is not in its docs (its pricing, read 27 October 2026). Compared with the others in desktop apps for coding agents.
• cmux is a macOS terminal for agents; on Windows and Linux see cmux on Windows.
• Herdr runs inside your own terminal on all three systems: Herdr vs cmux, and Orca vs Herdr side by side.
• T3 Code is a chat-style app for the same agents, with iOS and Android apps and several machines in one list: T3 Code vs Caprock.

What Caprock does not do

• Not the cloud, not CI. Sessions that run in a vendor's cloud or inside CI are not observed; Caprock sees the machines it is installed on.
• Not a bill. Every dollar Caprock shows is an estimate at API list price, as Claude Code's own /usage is: not a bill, and on a flat plan not money saved.
• No code editor. Files open read-only in a tab; you edit in your own editor.
• No SSH remotes, no cloud machines. It reaches only your other computers that run Caprock, through its relay.
• No native phone app. The phone uses its browser, which can be added to the home screen.
• No code-signed Windows installer yet. First launch on Windows needs SmartScreen's Run anyway; the Mac app is notarized.
• No cost for Grok or custom agents. Caprock starts them but does not read their turns.

Which to pick

• Orca — you want the fullest free cockpit: an editor, a browser, SSH hosts, native phone apps, any CLI agent.
• Caprock — you want the ledger: every session on the machine, who is waiting on you, how close each limit is, and what each pull request cost, with no telemetry. Keep Orca or your own terminal as the cockpit if you like. Install it.