aimonitor

Multi-account Claude Code usage monitor & silent account switcher for macOS & Linux.

English | 简体中文 | 繁體中文 | Tiếng Việt

Features

• 🔍 Live 5h + 7d usage bars per account — polled from Anthropic's /api/oauth/usage (no tokens consumed), with a trend line (↗ +21% in 45m).
• 🔀 Silent switching — aimonitor switch <label> refreshes the OAuth token and swaps the live credential. No claude /login, no terminal hop.
• 🤖 Auto-swap at the 5h or 7d threshold (default 80 %) — picks the account with the most overall headroom, skips exhausted/rate-limited ones, and rescues immediately if the active account hits 100 %. Running claude sessions follow automatically.
• 🔔 Threshold notifications as an account nears its limit (when auto-swap is off).
• 💾 Export / import settings, or migrate accounts to another machine — credentials optional and passphrase-encrypted (Argon2id + AES-256-GCM).
• 🔌 MCP server — Slack, ClickUp & Sentry tools for Claude Code over stdio, with per-service read-only mode.
• 🔐 OS-keyring storage (macOS Keychain, Linux libsecret). SQLite holds references; tokens never leave the keyring. No telemetry.

Install

# macOS (Sonoma 14+) — one command: taps, trusts, installs, clears Gatekeeper
curl -fsSL https://raw.githubusercontent.com/japananh/aimonitor/main/packaging/macos/install.sh | bash

# Linux (Ubuntu 22.04+) — CLI only
curl -fsSL https://raw.githubusercontent.com/japananh/aimonitor/main/packaging/linux/install.sh | sh

# Any platform, CLI only
go install github.com/japananh/aimonitor/cmd/aimonitor@latest

Prefer Homebrew directly? brew trust japananh/tap && brew install --cask japananh/tap/aimonitor, then clear Gatekeeper on first launch: xattr -dr com.apple.quarantine /Applications/AIMonitor.app (or right-click → Open). The one-line installer above does both for you. See docs/unsigned-app.md.

Upgrade

brew upgrade --cask aimonitor   # macOS
aimonitor update check          # CLI: is a newer release out?
aimonitor update install        # CLI: upgrade in the background

⚠️ macOS: brew upgrade --cask re-quarantines the app. Each upgrade lays down a fresh unsigned .app, so if it won't open afterward, re-clear Gatekeeper with xattr -dr com.apple.quarantine /Applications/AIMonitor.app (re-running the one-line installer does it for you). This is expected — not a broken install, so don't reinstall or --purge. The CLI keeps working regardless.

The menu-bar app also checks GitHub on launch and offers the update under Preferences → Check for updates. Pre-releases are never auto-served — brew upgrade keeps you on the latest stable.

Quick start

aimonitor add --adopt-current --label personal   # register your current Claude login
aimonitor add --label work                        # add another (drives claude /login, polls keychain)
aimonitor switch work                             # switch silently
aimonitor list                                    # live 5h / 7d usage per account
aimonitor doctor                                  # health check

Already on another switcher? aimonitor import pulls its accounts in one step. Auto-swap is on by default at 80 % — nothing else to configure for the common case.

Configuration

aimonitor config set auto_swap.enabled true        # default true
aimonitor config set auto_swap.threshold_pct 80    # 5h threshold
aimonitor config set auto_swap.threshold_7d_pct 80 # 7d threshold
aimonitor config set autostart true                # daemon at login

Back up or move to another machine:

aimonitor config export --out backup.json                                          # settings only (no secrets)
AIMONITOR_PASSPHRASE=… aimonitor config export --include-tokens --out full.json     # + encrypted credentials
AIMONITOR_PASSPHRASE=… aimonitor config import full.json                            # restore elsewhere

--include-tokens bundles your logins encrypted under the passphrase — restoring it means claude works on the other machine without re-login, so treat that file like a password. Same actions live in Preferences → Backup.

All config keys

Key	Default	Description
auto_swap.enabled	true	Master toggle for auto-swap
auto_swap.threshold_pct	80	5h utilization (%) to auto-swap
auto_swap.threshold_7d_pct	80	7d utilization (%) to auto-swap
auto_swap.grace_sec	60	Delay between the "pending" notification and the swap (0 = immediate)
notify.enabled	true	Warn as the active account nears its limit (only when auto-swap is off)
notify.warn_pct / notify.crit_pct	80 / 95	Warning / critical notification levels
daily_summary.enabled	true	Once-a-day notification recapping yesterday's token usage across accounts
auto_update.enabled	true	Check GitHub for releases on launch (never auto-installs)
autostart	false	Start the daemon at login
mcp.slack.enabled / mcp.clickup.enabled / mcp.sentry.enabled	true	Expose that service's MCP tools
mcp.slack.read_only / mcp.clickup.read_only / mcp.sentry.read_only	false	Hide the service's write tools
mcp.sentry.org	(empty)	Sentry organization slug the tools query
mcp.sentry.base_url	https://sentry.io	Sentry API host (set to your self-hosted host)
mcp.disabled_tools	(empty)	Comma-separated tool names to hide

How it works

The daemon polls /api/oauth/usage (~5 min ± jitter, no tokens consumed). When the active account crosses its 5h or 7d threshold, it picks the account with the most overall headroom, refreshes that account's OAuth token (POST .../v1/oauth/token), and writes it to the live Keychain slot. Running and new claude sessions adopt the new account — no /login, no restart.

See docs/architecture.md and docs/thresholds.md for the full picture.

MCP server (Slack, ClickUp & Sentry for Claude Code)

One stdio process serving Slack, ClickUp & Sentry tools — no extra runtimes.

aimonitor mcp connect slack     # store a Slack user token (xoxp-…)
aimonitor mcp connect clickup   # store a ClickUp token (pk_…)
aimonitor mcp connect sentry    # store a Sentry auth token (set mcp.sentry.org first)
aimonitor mcp register          # add the server to Claude Code

• Slack: post to channels/threads (mrkdwn, code blocks), upload, search, history, permalinks.
• ClickUp: workspace hierarchy, tasks, comments, attachments, Docs (read & write).
• Sentry: triage digests (projects, issue search, detail), root-cause (latest event stacktrace, tag/value distributions), and actions (resolve/ignore/assign, comment). Org-scoped and self-hosted-aware (mcp.sentry.org, mcp.sentry.base_url).
• Safety: Claude Code's per-tool prompts are the approval layer; per-service Enabled / Read-only switches and a per-tool disable list refine it. Tokens are verified live, then stored in the OS keyring — never in SQLite or logs.

Slack token scopes. The Slack token is a user token (xoxp-…). Grant these User Token Scopes on your Slack app (api.slack.com → OAuth & Permissions), reinstall, then connect — a missing one surfaces as slack: missing scope "…" on the affected tool: search:read, users:read, channels:history, groups:history, im:history, mpim:history, channels:read, groups:read, im:read, mpim:read, chat:write, files:read, files:write.

Sentry token scopes. A Sentry auth token (User Auth Token, or an Internal Integration). Set mcp.sentry.org — and mcp.sentry.base_url for self-hosted — before connect. Scopes: org:read, project:read, event:read for the read/triage tools; add event:write + member:read to resolve/assign/comment; event:admin to delete comments.

Privacy & security

• No telemetry, no phone-home. OAuth tokens live only in the OS keyring; SQLite holds references. Token bytes are never logged.
• Outbound traffic is limited to: GET /api/oauth/usage (introspection, no tokens consumed), POST /v1/oauth/token (silent token refresh), and the GitHub release check. Nothing about you is sent.

See docs/security.md for the threat model.

Troubleshooting

aimonitor doctor   # health check: config, SQLite, keyring, accounts

• "Daemon not running" / usage looks stale. Start or restart the background daemon with aimonitor config set autostart true, or click Start daemon in the popover — it registers a LaunchAgent that relaunches at login.
• App won't open — "damaged", "can't be opened", or "unidentified developer" (macOS Gatekeeper on the unsigned app). Happens on first launch and after every brew upgrade/reinstall — each lays down a fresh unsigned .app. Clear the quarantine flag, then reopen: xattr -dr com.apple.quarantine /Applications/AIMonitor.app. This is not a broken install, so don't uninstall or aimonitor uninstall --purge to "reset" it — --purge wipes your saved account logins. The CLI (aimonitor …) is unaffected. Details: docs/unsigned-app.md.
• Logs. The daemon writes to ~/Library/Logs/aimonitor/aimonitor.daemon.log (INFO/WARN/ERROR — never token bytes); background upgrades log to update.log beside it.
• Recent switches. aimonitor log prints the switch audit trail.

Uninstall

# Remove the app + daemon, KEEP your saved accounts
brew uninstall --cask aimonitor

# Full wipe — ALSO deletes every account login aimonitor saved in your Keychain.
# Purge runs first: it needs the binary still installed to reach the stashes.
aimonitor uninstall --purge && brew uninstall --cask aimonitor

🔴 --purge is destructive and irreversible. It permanently deletes aimonitor's saved copies of all your accounts (the aimonitor-<uuid> Keychain stashes) — you'll have to aimonitor add and complete /login for every account again. Plain brew uninstall --cask aimonitor (no --purge) keeps them all.

App won't open? That's Gatekeeper, not a broken install — don't --purge to "reset" it. Run the xattr fix in Troubleshooting instead.

--purge never touches Claude's own Claude Code-credentials slot, so the account you're currently signed into keeps working in claude with no re-login — only aimonitor's saved copies of your other accounts are dropped (and those you must re-add).

Build from source

Requires Go 1.25+. Pure Go (CGO_ENABLED=0 works on macOS; keychain access shells out to /usr/bin/security).

make build              # CLI binary
make test               # unit tests
make widget             # AIMonitor.app (macOS; needs the Swift toolchain)
make release-snapshot   # goreleaser dry-run

License

MIT © @japananh