Many of us now keep more than one coding agent: Codex for one workflow, Claude Code for another, maybe Grok, OpenCode, dsh, or Antigravity on the side. Usage lives in different UIs — five-hour windows next to weekly caps next to “All models” labels — and cost language mixes subscriptions, client estimates, and API list prices until nothing is comparable.
Agent Usage is a macOS-first, fully local multi-agent usage center. One command opens a dashboard for native quota windows and reset times, tokens and model rankings, history/diagnostics, and a public API retail equivalent for eligible usage. It offers advice only and never switches agents automatically.
- Repo: https://github.com/Blackman99/agent-usage-all-in-one
- Compare docs: https://github.com/Blackman99/agent-usage-all-in-one/blob/main/docs/comparison.md
- npm:
agent-usage-all-in-one· CLI:agent-usage· MIT - Requires macOS + Node.js 24+
- No Cursor coverage yet · don’t read the package name as “every agent on earth”
Quick start
npx agent-usage-all-in-one
agent-usage The API retail equivalent is **not a bill**. Unknown prices stay **unpriced** — never guessed as `$0`. **Grok Build/SuperGrok** and **xAI API** are independent billing domains and are **never summed**.
### Light comparison
Peers worth naming honestly (see docs/named-competitors.md):
- **ccusage** — CLI reports; we lead with Dashboard + native quotas
- **CodeBurn** — multi-tool/multi-platform; we lean stricter cost semantics + default no telemetry
- **tokscale** — token/cost scale peer; same honesty boundaries apply
- **Token Monitor** — closest dashboard shape; we do **not** claim Cursor / multi-machine / menu bar yet
### Feedback asks
1. Which connector should deepen first?
2. Does retail-equivalent framing help or confuse?
3. What would make `start-at-login` a habit?
PRs/issues welcome. Don’t paste Keychain secrets or live exports.
Prefer npx; global install optional. Daemon binds 127.0.0.1. Data under ~/Library/Application Support/Agent Usage.
Cost honesty
Actual charges, provider estimates, fixed subscription fees, and API retail equivalents are separate evidence.
Top comments (2)
This is the clearest cost-semantics writeup I've seen for the multi-agent problem — I run several coding agents across a small VPS setup and the "subscription vs client estimate vs API list price" mixing is exactly where our internal numbers went wrong twice. Keeping unknown prices unpriced instead of defaulting to $0 is a detail most dashboards get wrong; it's what made me actually trust the retail-equivalent column.
Two practical questions from a mostly-headless setup: does it read the quota windows purely from local JSONL logs, and how fragile is that when sessions run under a different user (we hit log-path surprises on our boxes)? And is a Linux/no-GUI mode on the roadmap — macOS-first is a real blocker for anyone whose agents live on a server.
Some comments may only be visible to logged-in visitors. Sign in to view all comments.