Limits and Usage
2 min read
The generator has fixed ceilings. They are published here rather than discovered, so you can plan a run around them.
Per run
Section titled “Per run”| Limit | |
|---|---|
| Requirements in one brief | 10 |
| Documents attached | 3 |
| PDFs, of those documents | 1, up to 8 MB |
| Scenarios mapped | 40 (default 30) |
| Test cases written per run | 20, in batches of 10 |
| Test cases per run, from a requirement | 10 |
Nothing here is a soft limit. If you need more, split the work: a run per feature area usually produces a better map than one run spanning everything, because the scenario budget is shared across whatever you supply.
Source length
Section titled “Source length”Each source is capped at 20,000 characters, and the whole brief at 60,000.
A source that is too long is trimmed from the middle, keeping the start and the end. That is deliberate: acceptance criteria cluster at the end of a specification, so trimming the tail would throw away the most useful part.
When it happens, the run says so and names the source with the number of characters kept out of the total. You are never left guessing whether the whole document was read.
The workspace AI allowance
Section titled “The workspace AI allowance”AI generation is included rather than metered, so Hawzu shows an allowance percentage rather than a token count or a bill.
You can see it in:
- Workspace Settings, under general settings,
- the profile menu in the workspace and project headers,
- the project’s left rail,
- the generation modals themselves, while you work.
The allowance covers a calendar month and resets at the start of the next one. When it runs low the wording changes to say so; when it is exhausted, generation is unavailable until the reset and every other feature carries on as normal.
Two things spend it:
- Each generation — one coverage map is one generation; each batch of ten written test cases is another. Writing 20 cases is therefore three generations, not one.
- Attached PDFs — a PDF costs more than the same content as text, because the document is read as a document rather than flattened.
When a run does less than you asked
Section titled “When a run does less than you asked”The generator is built to under-deliver rather than to invent. Three cases where it deliberately returns less:
- Fewer scenarios than requested — the source didn’t support the number you asked for. The map says so instead of padding the list.
- Fewer test cases than scenarios ticked — the same, one stage later.
- Nothing at all — the source was too thin to write a single grounded case, and it tells you rather than producing something plausible.
A short honest map beats forty plausible titles you then have to disprove one at a time.