Credits and usage
WeHarness counts the work it does for your workspace in credits. You see credits, never a price in a currency. Open Usage & billing in the sidebar to see what is left, what was added, what this month used and who used it.
Getting credits
Section titled “Getting credits”During the pilot nothing is bought in the app. WeHarness adds credits to your workspace, usually as a package. A new workspace starts with none. To add credits, write to info@weharness.io.
| Package | Credits | Includes |
|---|---|---|
| Starter | 1,500 | — |
| Medium | 2,500 | Bring your own key for Gemini, OpenAI and Claude models |
| Large | 5,000 | Bring your own key for Gemini, OpenAI and Claude models |
Usage & billing lists these packages as coming soon: buying one arrives with payments. Until then WeHarness assigns a package to your workspace on request.
Proposed, not purchasable
Section titled “Proposed, not purchasable”Usage & billing also shows what is proposed for when payments arrive. None of it can be bought or is charged today.
- Team (10,500 credits) and Scale (22,500 credits) packages, with bring your own key.
- Top-ups of 500 or 1,000 credits, added to an active package. They do not change what the package includes. Automatic top-up stays off.
- A trial for public sign-up: one trial of 100 credits per verified person, for one workspace, with an expiry.
- Monthly renewal: a package’s credits renew each billing month and unused ones do not carry over; top-ups stay valid for 12 months; work already reserved keeps its credits across a renewal.
What uses credits
Section titled “What uses credits”| Operation | Credits |
|---|---|
| A session message WeHarness answers | By what its models used: at least 1, at most 100. |
| A plan version WeHarness produces | By what planning used: at least 5 (the planning fee), at most 25. On your own model key: 5. A plan that only asks you questions: none. |
| A decision between options (Elephant-1) | 1. |
| Preparing an approved environment | 1 for each new build. Reusing a prepared environment, or a build that fails, uses none. |
| Cloud work: sandboxes, previews, evaluation runs and the models your code calls | By what it used, never more than it reserved before it started. |
| Judges that grade an evaluation | By what their calls used, at the judge model’s price, within the evaluation’s judges line. A failed run’s judges use none. |
| An online evaluator scoring a trace | At least 1 for a judge or decision, within the application’s daily cap. Code evaluators use none. |
| Traces your applications send | 10,000 spans a month are included, then 1 credit per 1,000 spans, counted over the month. |
| Reading the catalog, plans, stored results, traces and status | None. |
| An operation that failed without a useful result | None. WeHarness bears that cost. |
| A model call paid by your own API key (Model keys) | None. A message answered entirely on your key uses none at all. Sandbox time, decisions and a plan’s planning fee still count. You are never charged twice for inference your key paid. |
Each operation is counted once, however often WeHarness retries it behind the scenes. The record of credits only grows: a correction is a new entry, never an edit.
Only models with a price are offered, so every model call can be counted. See Models.
Every run has a cap
Section titled “Every run has a cap”| Run | Cap |
|---|---|
| A message | 100 credits |
| A plan version | 25 credits |
| An evaluation | Its judges line, shown before it starts |
| Improve this agent | Its total cap, shown by part before it starts. See Improve this agent. |
A member’s per-run budget can lower these caps (see below).
When a message reaches 80% of its cap, WeHarness stops taking new steps and writes its reply with what it has. If the reply itself would go past the cap, it stops and says This message reached its credit cap. A plan or a judge that reaches its cap stops visibly too. Nothing is ever charged above a cap. Work on your own model key counts toward a cap at the same rate.
Budgets for members
Section titled “Budgets for members”Owners and admins can set a budget in Usage & billing, for everyone or for one person:
- credits per month;
- the most one message, plan or check may use (a per-run budget).
Work on the workspace’s own model keys counts toward a budget at the same rate, because the workspace’s owner pays the provider for it.
- When a member’s monthly budget is reached, their new work waits:
402 MEMBER_BUDGET_REACHED. Work already running finishes. - Cloud work that would reserve more than the person’s per-run budget does not start:
402 RUN_BUDGET. - Usage & billing shows each person’s use this month. Members see only their own line.
- Viewers never start work that uses credits.
Traces
Section titled “Traces”Usage & billing shows this month’s accepted spans against the included 10,000. A span is one step of a trace, not a whole trace, and each is counted once: a retried delivery is not counted again. Above the allowance a workspace uses 1 credit per 1,000 spans, counted over the month rather than per delivery, so the month’s total is rounded once. Each day’s spans are settled the next day; the page shows what this month’s spans come to so far.
Cloud work reserves its credits first
Section titled “Cloud work reserves its credits first”Before cloud work starts, WeHarness reserves the most it can use, so work you approved is never stopped halfway for credits. Available credits are what is left less what running work has reserved. Usage & billing shows both.
| Cloud work | What it reserves |
|---|---|
| A command in a session | One working sandbox: its time limit and the models your code may call in it |
| A clean check | One check sandbox |
| A preview of your app | One preview sandbox |
| An evaluation run | One evaluation sandbox, and its judges: each judgment its cases ask for, at the judge model’s price |
| A replay on recorded traces | Its judges only: no code runs |
| Approving an environment | Its setup: the 1-credit build fee |
| A task WeHarness works on in the cloud | The check of its report first, then its working sandboxes and two repair rounds |
The phases are shown separately: setup, the working sandbox, checks, repair rounds, previews, evaluations and judges. Usage & billing lists what each kind of work reserves today.
- When the available credits cannot cover the reservation, the work does not start and nothing of it is recorded. Work already running finishes.
- When the work ends, it is debited only what it used, and the rest is released at once. If a provider bills more than was reserved, WeHarness bears the difference.
- A reservation whose work is still running is renewed before its deadline, within the credits it reserved, at most 3 times.
- If your code’s model calls used most of a sandbox’s model allowance, the next sandboxes of that task or session reserve more for them, up to four times the usual, and the session says so. When a task’s approved work cannot cover its next round, WeHarness asks for an extension first: an owner or admin approves more credits for that session in the web app. WeHarness never switches to a weaker model to fit a budget, and never charges more on its own.
- The models your code calls in a sandbox stop answering once that sandbox’s share is used.
When WeHarness is busy
Section titled “When WeHarness is busy”New messages, plans, checks and evaluations can wait a minute when:
- your workspace already has many pieces of work running (
429 WORKSPACE_BUSY); - your workspace already uses its two sandboxes for people’s work (during the pilot, two at a time per workspace; checks have their own room);
- WeHarness is at capacity (
503 AT_CAPACITY), or one model’s provider is (503 PROVIDER_BUSY).
The answer says when to try again. When WeHarness’s queue falls behind, the message box says WeHarness is busy before a message is refused. Reading sessions, plans and status, and work already running, are never held back. Work WeHarness has accepted keeps its place in the queue and finishes. WeHarness never switches your session’s model by itself.
When credits run out
Section titled “When credits run out”Usage & billing warns when less than a tenth of the credits added is left. When none are left (402 OUT_OF_CREDITS):
- Work already running finishes, even if it uses more than was left. The next credits added cover the difference.
- New work waits until credits are added: messages, plans, decisions, checks and evaluation runs. A coding tool that asks over MCP gets the same answer.
- Reading your sessions, plans and status, pushing finished work and exporting your data still work.
WeHarness never switches to a lesser model or skips a check because credits run low.
In a session
Section titled “In a session”The line under a session’s work panel shows the credits that session used so far, with its model calls and sandbox time, and what its running cloud work has reserved.