Skip to content
ReviewRamble

The weekly brief

AI, reviewed, in your inbox.

  • New reviews of tools, models, MCP servers and repos, with the verdict up front
  • Re-checked prices and ratings when something changes
  • New analysis pieces as they are published

Models, agents, MCP servers and the tools actually worth your time.

One email a week. Unsubscribe with the link in any brief. How we use your email

Caprock

Local desktop app that tracks Claude Code, Codex and OpenCode sessions in one window, with estimated spend per repo and plan-limit alerts.

Reviewed by ReviewRamble editors · Updated

View on GitHub Free plan

Share this page:XLinkedInFacebookRedditWhatsAppEmail

Our rating

3.6/5

Caprock tracks coding-agent sessions, limits and estimated spend in more detail than most rivals, all on your own machine, but it is three months old, moves at several releases a day and asks for a lot of trust in a small maker.

What is Caprock?

Caprock is a desktop app for macOS, Windows and Linux, built in Go with a Tauri shell, that puts the coding-agent sessions on your machine into one view. It follows Claude Code through hooks and transcripts, reads Codex transcripts, OpenCode and Hermes Agent databases (read-only), Cursor chats and Gemini CLI telemetry, and can start most of them in its own terminal tabs and split panes, one per project or git worktree. Sessions waiting on a permission prompt are listed first and can be answered from a notification, the menu bar or a paired phone.

Cost is an estimate: every turn is priced at API list prices and grouped by repository, monorepo service and pull request, and Claude's 5-hour and weekly windows are read through Caprock's status line, with an alert at 90%. An optional task runner, off by default, gives each worker its own worktree under .caprock-worktrees on a caprock/ branch and never merges. A read-only MCP server and Claude Code plugin let agents query past sessions, notes, cost and CI status.

The code is Apache-2.0 on GitHub under dspv/caprock, published by Cybrix LLC of Wyoming. A local daemon keeps everything in one SQLite file and serves the dashboard on localhost:22776. The pricing page says 83 of its 91 features are free; Premium at $30 a year or $100 once adds a daily budget auto-pause, Telegram replies and more than three accounts per agent.

Is Caprock worth it?

Caprock solves a bookkeeping problem that grows with every agent you add: which session is stuck on a prompt, which repo the tokens went to, and how close each Claude or Codex account is to its limit. Most tools in this corner do one of those; Caprock tries to do all of them from local files, with no account.

The privacy design is careful on paper. The README says nothing leaves the machine unless you switch it on, the release check, GitHub, Telegram, Gemini and phone pairing are each off by default, and the licence key is checked locally. The phone relay is end-to-end encrypted, and the privacy policy says the relay stores nothing. Only the website uses Umami analytics, without cookies.

The spend figures need reading correctly. They are API list-price estimates, not a bill, so on a flat Claude Max or ChatGPT plan they show what the work would have cost, not what you paid. Codex turns with no model named show no cost, and OpenCode, Hermes and Cursor figures are passed through as those tools report them.

The risks sit in the defaults you opt into. The SQLite file holds prompts, replies and tool output, which the README itself warns is where secrets end up; it is kept at 0600 permissions. Task-runner workers run with --dangerously-skip-permissions, the Windows installer is not code-signed yet, and the agent install route pipes a script from caprock.dev into a shell. A --lan flag exposes the dashboard to your network after pairing.

Maturity is the open question. The repository was created in August 2026, had 50 stars and 3 forks on 20 November 2026, and shipped v0.184.2, v0.185.0 and v0.185.1 within two days, so behaviour can change under you. Claude Squad is far narrower but has nearly 8,700 stars; Caprock has the features, not yet the track record.

Where it earns its place

  • A freelance developer with two Claude accounts and one Codex account

    Stop hitting a usage limit mid-task without warning

    Caprock shows each account's 5-hour and weekly windows side by side and alerts at 90%; a session that hits Claude's limit can continue on another account. Up to three accounts per agent are free; more need Premium at $30 a year.

  • A tech lead at a 10-person startup paying for agents by API

    See which repo and pull request the agent spend went to

    The pipeline view lists pull requests with their CI checks and what the agents spent on each, and caprock report produces a shareable cost summary. Figures are list-price estimates, which match a pay-as-you-go API bill more closely than a flat plan.

  • A developer who leaves agents running while away from the desk

    Answer permission prompts from a phone

    Pair a phone by QR code, either the free iPhone app or a browser, and approve or deny prompts over the encrypted relay or local Wi-Fi. The relay sees which machines talk and when, but not the content, according to the docs.

Before you commit

Setup
Install with brew install --cask dspv/tap/caprock-app, a notarised .dmg, a Windows installer or an AppImage, .deb or .rpm; on first run it asks before adding a hook to ~/.claude/settings.json and backs the file up. A daemon-only install runs headless over SSH.
Integrations
Claude Code, Codex, OpenCode, Gemini CLI, Hermes Agent, Cursor and DeepSeek; GitHub for pull requests and CI; Telegram and an iPhone app for alerts; a read-only MCP server via caprock mcp; export to TSV, CSV or JSONL.
Data and privacy
One SQLite file on your disk holding prompts, replies and tool output, at 0600 permissions; the dashboard listens on 127.0.0.1 port 22776. The README states no account and no telemetry. Bug reports and surveys go to caprock.dev only when you press Send.
Cost at scale
Free for individuals; Team is a flat $6,000 a year, or $500 a month with a $500 setup fee, self-hosted, with only usage totals leaving each laptop.
Cancelling and refunds
The yearly Premium plan renews until cancelled and stays active to the end of the paid period; an expired key keeps Premium working for seven days. The $100 option never renews. Refunds are offered only for the affected period if a paid feature is removed or materially unavailable.
Limits
Gemini CLI is tracked only for sessions Caprock starts; Codex costs come from Caprock's own price table and some turns are upper bounds; the task runner stops and asks after three failed rounds of done checks, each limited to five minutes.

What’s good

  • No account and no telemetry; network features are opt-in one by one
  • Reads many agents, including sessions started outside the app
  • Spend per repo, service and pull request, plus plan-limit alerts
  • Apache-2.0 code with a read-only MCP server and Claude Code plugin

What’s not

  • Created in August 2026, 50 stars, very rapid release churn
  • Costs are list-price estimates, not what a flat plan actually charges
  • Local database stores tool output, which can include secrets
  • Unsigned Windows installer; task runner skips agent permission prompts

Best for

Developers running several Claude Code and Codex accounts in parallel who want a local view of stuck sessions, limits and estimated spend per repo, and can tolerate a young project.

Caprock alternatives

Caprock: common questions

Is Caprock worth it?
Caprock tracks coding-agent sessions, limits and estimated spend in more detail than most rivals, all on your own machine, but it is three months old, moves at several releases a day and asks for a lot of trust in a small maker. We rate it 3.6 out of 5.
How much does Caprock cost?
Community edition free and Apache-2.0 (83 of 91 features, no account or key). Premium $30/year or $100 once adds 8 features such as daily budget auto-pause, plan-limit pause, Telegram replies and more than three accounts per agent. Team $6,000/year or $500/month, self-hosted in your own infrastructure; Enterprise by quote.
What are the downsides of Caprock?
Created in August 2026, 50 stars, very rapid release churn; costs are list-price estimates, not what a flat plan actually charges; local database stores tool output, which can include secrets; unsigned Windows installer; task runner skips agent permission prompts.
Who is Caprock best for?
Developers running several Claude Code and Codex accounts in parallel who want a local view of stuck sessions, limits and estimated spend per repo, and can tolerate a young project.
How much does Caprock cost for a team or in production?
Free for individuals; Team is a flat $6,000 a year, or $500 a month with a $500 setup fee, self-hosted, with only usage totals leaving each laptop.
Is Caprock safe for company data?
One SQLite file on your disk holding prompts, replies and tool output, at 0600 permissions; the dashboard listens on 127.0.0.1 port 22776. The README states no account and no telemetry. Bug reports and surveys go to caprock.dev only when you press Send.
How do I cancel Caprock, and can I get a refund?
The yearly Premium plan renews until cancelled and stays active to the end of the paid period; an expired key keeps Premium working for seven days. The $100 option never renews. Refunds are offered only for the affected period if a paid feature is removed or materially unavailable.
What are the best Caprock alternatives?
The closest alternatives we have reviewed are Claude Squad, Vibe Kanban, ccusage and Orca.

Answers checked Nov 20, 2026.