harnsy

harnsy Max

See the limit before you hit it.

Usage shows every 5-hour and weekly limit of your logins with a forecast to the reset, the history of each limit, and what your agents spent. It’s in every harnsy build; a harnsy Max key turns it on.

Limitsharnsy Max

Each login: subscriptions from Connections and each harness’s own login. The bar is used now; the pale part is the forecast to the reset; the thin mark is an even pace.

Switch candidates

Personal · week 31% now, about 48% by the reset

Switch a project’s subscription
Usage in the dashboard: each subscription’s limits, and whether they last until the reset. Sample data.

Limits

Every limit in sight.

For each login, the 5-hour and the weekly window as bars: the bar is what you used, the pale part is the forecast to the reset, a thin mark is an even pace, and the reset time is beside it.

A verdict for each.

Enough, tight, or runs out at a time. It also says when a limit is used up until a time, when there is no fresh reading, when it reset since the last reading, or when there are too few readings.

How the forecast works.

It carries on the recent pace: the last hour for the 5-hour window, the last day for the week, and the whole window when there are fewer than three readings in that span (it needs three readings over at least five minutes). “Tight” starts at 90%, “runs out” at 100%. It is an estimate, not a promise.

Two logins, one subscription.

Mark a connection “Same subscription as” another and the two count as one account.

Why a limit is missing.

When the limits of a Claude login cannot be shown, a notice says why instead of leaving an empty bar.

Spend

Codex input and cache reads, separately.

Codex input tokens and cache reads are counted separately in the spend table.

A way out.

When a login runs tight or out, Usage suggests another subscription to move a project to, with “X% now, about Y% by the reset”, and a link to switch it in Connections.

The history of every limit.

Charts of the limit windows over the last 24 hours or 7 days, the limit dashed. History starts when you install this version; nothing is filled in backwards.

Connections →

Where the tokens go.

Spend for today, 7 or 30 days, grouped by project, team, role, agent, login, harness or day: cost, then input, cache write, cache read and output tokens, then sessions, with a total. Cost is the harness’s own figure at API prices, not a bill.

Which agent eats the week.

Group the spend by agent or project to see who uses most, before the weekly limit is gone.

A daily budget (experimental).

Off until you switch it on. It shows how much of each weekly limit may go today so the limit lasts to the reset, as a “today” bar under each weekly bar. Leads can read it before they hand out heavy work.

A daily allowance, hour by hour.

harnsy Max

With harnsy Max, the day's share of a subscription's limit is spread over your working hours, and the spend is shown by model and by hour. In the evening the allowance no longer drops to almost nothing. The lead is warned ahead of the limit and stops softly before 100 %.

Where it comes from

Taken from your machine.

The numbers are read once a minute on your machine and stay there. Claude Code: its own status-line snapshot for the limits and cost, and the session transcripts for tokens. Codex: the limit windows and token totals in its session files. OpenCode: cost and tokens per message.

What eats the tokensFindings: 3
  • acme-leadContext per turn312kLeaking
  • acme-frontTurns without action41%Worth a look
  • acme-reviewWakes a day23Better

Review the spend

What eats the tokens: one finding per role, with its level.

See what eats the tokens.

“What eats the tokens” on a team’s spend tab shows, per role, the habits that waste context, such as repeated re-reading and wake-ups, with advice. “Review the spend” hands the lead a review: roles propose lessons for their post, and you accept the role changes.

Two logins, kept apart.

A personal Claude login and a corporate subscription token no longer merge into one row: both stay. “Not the same” ends the suggestion for good and splits a pair that was merged.

Spendharnsy Max

What the agents spent, from their sessions’ running totals. History starts when this version was installed.

Period:today7 days30 daysGroup by:projectteamroleagentloginharnessday
projectCostInputCache writeCache readOutputSessions
acme-shop$1321.2M3.4M41.8M610k38
acme-api$58.40540k1.5M17.2M260k21
docs-site$12.10120k330k3.9M70k9
no project$1.8518k40k610k9k3
Total$2041.88M5.27M63.5M949k71

Cost is the harness’s own figure at API prices, not a bill; subscriptions report it too. Codex reports tokens only.

Spend by project: tokens, sessions and cost at API prices. Sample data.
Spendharnsy Max

What the agents spent, from their sessions’ running totals. History starts when this version was installed.

Period:today7 days30 daysGroup by:projectteamroleagentloginharnessday
agentCostInputCache writeCache readOutputSessions
acme-shop-backend$61.40480k1.4M18.2M260k14
acme-shop-lead$34.20260k700k9.8M140k11
acme-shop-frontend$22.75190k520k6.1M98k8
acme-shop-tester—310k—4.2M120k6
acme-shop-reviewer$13.85120k310k3.6M64k5
Total≥ $1321.36M2.93M41.9M682k44

Cost is the harness’s own figure at API prices, not a bill; subscriptions report it too. Codex reports tokens only. Some sessions report tokens only (Codex): the sum is partial.

Spend by agent over 7 days: who uses up the week. Sample data.
Limitsharnsy Max

Each login: subscriptions from Connections and each harness’s own login. The bar is used now; the pale part is the forecast to the reset; the thin mark is an even pace.

History24 hours7 daysdashed: the limit
A limit’s history over 7 days, the limit dashed. History starts when you install this version. Sample data.

A key turns it on.

Without a key, Usage shows a short note about harnsy Max instead. Add a harnsy Max key on the License page to turn it on.

Apply for a key →

What's in harnsy, and what's in harnsy Max? →

Limits, honestly

  • Codex reports tokens only, no money, so sums that include Codex sessions are partial.
  • OpenCode reports no limit windows: it has spend, but no limit bars and no weekly budget.
  • Cost is counted at API prices; what a subscription really costs you is set by its vendor.
  • A forecast needs a few readings: a new login shows “too few readings” at first.

How to set it up →

Your first team is one prompt away.

Installing takes one sentence to your agent. To run your agents, harnsy drives a terminal: tmux if it is installed, otherwise its own terminal host (on Windows, always its own).

Install harnsy following https://harnsy.dev/llms.txt

Questions

What's in harnsy, and what's in harnsy Max?

harnsy itself is free: 1 organization, 2 projects, no limit on agents, one machine. Claude Code, Codex, OpenCode and pi need their own plans or keys, paid to their providers. harnsy Max is the same build with a key: more projects and organizations, access from your phone (beta), several machines as one team, several subscriptions and their usage.

All questions →
site-3flead · Claude Codeliveview only
❯ Plan #42 with the team.

Waiting for the breakdown from analyst-7a…

from analyst-7a through harnsy❯ #42 broken down: three acceptance criteria, including a retry after 24 h.

I’ll hand #42 to site-9a.

❯