Open source alternative to Conductor
Last verified
Track every coding agent session you run, with cost per repository, plan limits, and a PR pipeline view. Runs locally, no account required.
Caprock shows every active session, what is waiting on approval, your plan windows in real time, and per-repository cost in one window. The Pipeline view tracks each pull request from the first agent commit to the release it shipped in.
You run a single agent on one project and check usage from the provider's own dashboard. Caprock adds the most value when multiple sessions run across multiple repositories.
Caprock imports transcripts still on disk on first install, then records every new session going forward. After Claude Code deletes its own files, the sessions remain searchable in Caprock with their full tool call history.
You already back up transcripts manually or only need short-term session visibility within the default 30-day window.
A phone paired by QR code opens Caprock in its browser from anywhere through an encrypted relay. Sessions waiting on approval come first. You can dictate instructions, approve commands, and watch what an agent shipped without returning to your desk.
You rarely leave your desk while agents are running, or you prefer not to enable the phone relay network feature.
Cost per repository breaks down to per-service in a monorepo, each turn counted toward the directory of the last file the agent touched. caprock report prints a publishable summary at API list prices. caprock export writes the underlying data as TSV, CSV, or JSON for further analysis.
Your goal is aggregating cost across a team or integrating spend data into a billing system. Caprock is a solo tool and does not aggregate usage across multiple users.
Running multiple coding agents across several projects is hard to track. Claude Code deletes its own transcripts after 30 days by default, leaving no long-term record of what an agent did, what it cost, or which pull requests it produced. When a session waits on a permission prompt, nothing surfaces it. When an agent loops on the same tool call, the cost grows without a visible warning.
The cost side is equally opaque. Paid AI coding plans bundle flat fees with usage windows, but the underlying token consumption and what the same work would cost through the API are not surfaced by default. A 5-hour or weekly Claude window can run out before any warning appears. Nothing in the default tooling shows which repository or directory within a monorepo is responsible for the spend.
Claude Code deletes transcripts after 30 days by default. Caprock captures every session in a local SQLite database, including ones started in your own terminal. On the maker's machine, 67 of 114 transcripts were already deleted after installation, including 10 from deleted worktrees, and all 67 remained searchable in Caprock. New sessions are recorded as they run and kept indefinitely.
Every agent turn is priced at API list prices, broken down by repository. Within a monorepo, cost is allocated to the directory of the last file the agent touched, so the rows add up to the repository total exactly. This is an estimate of what the same work would cost through the API, not a bill. In August, the maker found he had run $5,898 of Claude Code on a $200 Max plan using this view.
Caprock shows your Claude 5-hour and weekly usage windows in real time, with the time each resets and a warning before you hit 100%. Codex windows appear beside them so you can switch agents before running out. The menu bar shows your plan limits, today's spend, and the sessions waiting for approval at a glance.
The Pipeline view shows every pull request with its CI checks per job and operating system, how long each check took, and what the agents spent on that branch. A failed run, a conflict, or a branch behind main surfaces an action from the pull request row. Releases list which pull requests shipped in each version with their agent cost summed.
One QR code scan pairs a phone that opens Caprock in its browser from anywhere, through an end-to-end encrypted relay that sees only ciphertext. Sessions waiting for your approval come first. You can dictate instructions, approve commands, and monitor what an agent shipped without sitting at your computer.
When a session calls the same tool with nearly the same input five times within three minutes, a banner goes up. Repeated file reads are excluded; on 64,733 real tool calls they were over half of all alerts. Sessions Caprock started can be paused automatically when a loop is detected.
Both tools serve developers running AI coding agents who want visibility beyond the default CLI experience. Conductor is a paid commercial tool in this category.
Caprock's approach is local-first and open source. Session data stays in a SQLite file on your disk; no account is required, and nothing leaves your machine unless you enable a specific network feature. The Apache-2.0 code is on GitHub. The features available without payment include live sessions, cost per repository at API list prices, Claude and Codex plan-limit windows, session history past the 30-day cleanup, session search, and the pull request pipeline from agent commit to release. Eight additional features (auto-pause at a daily budget, Telegram answers, weekly reports, and others) require a paid license at $30 per year or $100 once.
Caprock is the better fit for a solo developer who wants a permanent record of every agent session, per-repository cost visibility, and plan-limit warnings with no cloud account and no subscription for core functionality.
Caprock does not aggregate usage across team members or offer a managed cloud deployment, so a team that needs those should look at a paid tool such as Conductor.
Install the macOS desktop app with Homebrew.
```bash
brew install --cask dspv/tap/caprock-app
```By default, no. Data stays in a SQLite file on your disk, and Caprock listens only on 127.0.0.1. There is no account or telemetry. Network features (update checks, GitHub, phone pairing, Telegram) are each off until you turn them on. Updating downloads the new app from GitHub's release files; that is the only automatic network activity once you allow it.
Yes, if those transcripts are still on disk. Claude Code keeps transcripts for 30 days by default. Caprock imports everything still present on first run, then records new sessions going forward. On the maker's machine, 67 of 114 transcripts were already deleted when Caprock was installed, and all 67 remained searchable after import.
No. Keep running claude or codex in your own terminal. Caprock adds a small hook to ~/.claude/settings.json, backed up first, and reads the transcripts those agents already write. You can remove the hook with caprock hooks uninstall; your recorded session history stays intact.
Yes, for solo use. Live sessions, cost per repository, plan-limit warnings, history, and search are free and described as staying free. Eight additional features require a paid license: auto-pause at a daily budget, pause at the plan limit, Telegram answers, a weekly report, usage queries, failed-tool tracking, session export, and Obsidian import. A license costs $30 per year or $100 once.
Claude Code, Codex, OpenCode, Gemini CLI, Hermes Agent, and DeepSeek's agent. Each agent writes its own transcript format; Caprock reads all six into a single table on your disk. Sessions started by Caprock and sessions you start in your own terminal both appear in the same view.
Alternative to
Grok Bot
Self-hosted AI agent management with org charts and budgets
Alternative to
Conductor
Open-source ADE for Claude Code, Codex, and Cursor in one thread
Alternative toConductor
Desktop canvas for persistent multi-agent AI teams, Windows-only
Alternative toConductor
Agent runtime that keeps terminals alive across machines
Alternative toConductor
Run a company of AI agents on a server you own.
Alternative toConductor
Local-first terminal workspace for AI coding agents
Sponsors
Alternative to
Superhuman
AI email assistant that drafts replies and clears your inbox
Alternative toClaude Managed Agents
Open-source agent deployment API. Self-host or use Omnara Cloud.
Alternative toGrok Bot
AI teammates you own: your keys, your model, your machine.
The projects we add and every new guide. Free, unsubscribe any time.
Claude Code to 
OpenCode
Jira to 
Windshift
Notion to 
AppFlowy
Slack to 
Mattermost
Linear to 
GitLab
Wispr Flow to 
Handy
Arc to 
Zen Browser
CapCut to 
Concat
Grok Bot to 
rakazo
Superhuman to 
Inbox Zero
Airtable to 
AITable
Muse to 
OpenClaw
Railway to 
Coolify
GitHub Copilot to 
Cline
Datadog to 
Netdata
Microsoft Teams to 
Rocket.ChatConfluence to 
Docmost
Pinecone to 
Qdrant
Zapier to 
n8n
Devin AI to 
OpenHands
Firebase to 
Supabase
1Password to 
KeePassXC
Contentful to 
Strapi
Adobe Premiere Pro to 
Shotcut
