Limits

Three separate things get counted: companies served by Prospex, requests made to the API, and research performed inside a project.

Rows: a monthly allowance of companies per plan

A row is one company. Each company counts once a calendar month for the account, whichever surfaces served it and on however many days: the same company opened in the app, pulled from a feed and read through the API and MCP costs one row that month. A page of signals costs one row per company the signals are about.

• The daily ceiling is half the monthly allowance on every paid plan. It counts companies new to the month, so re-reading a list you already pulled this month is free.
• The day resets at midnight Europe/Zurich, and the month at midnight on the 1st.
• A plan change from the account page applies to both ceilings: an upgrade at once, a downgrade at the next renewal. The pricing page has the prices.

These surfaces take companies and spend rows:

Every JSON page carries a usage block with rows_today, daily_limit, rows_this_month and monthly_limit, so a workflow can stop on its own terms.

When the budget runs out partway through a page, the page carries the companies that fit, has_more is true, and the cursor resumes at the first company held back. When not one company of a page fits, the page is refused: 429 daily_row_limit with Retry-After pointing at midnight, or 429 monthly_row_limit with Retry-After pointing at the 1st. Companies already served today count nothing, so you can read them again at the ceiling.

A CSV response streams up to 10,000 rows in one go and stops where the budget does, so a single export of a large list can take the whole day. The budget is per account: all users and automations share it.

Browsing on the website: no rows spent

The tables on the website show companies without spending rows. They have their own ceiling instead, of distinct companies a day: 100 on the free plan and 1,000 on a paid one. A company already taken this month browses free. Past the ceiling, a table shows the companies that fit and says how many it held back until midnight, Swiss time.

The browse ceiling applies to the website only. Every feed, export, API and MCP read is a take and spends rows from the monthly allowance.

Requests: 10,000 per account per day

• Counted on the API only, and reported in X-RateLimit-Limit, X-RateLimit-Remaining and X-RateLimit-Reset on every response. Past it, 429 rate_limited.
• A request body is capped at 1 MiB. Larger is refused before it is parsed.
• One POST .../prospects carries at most 100 companies.
• A list page is at most 500 rows and a project page at most 200. Ask for more and you get the ceiling.
• Events stay replayable for 30 days. Past that a webhook cannot be replayed, and the state it described is still readable through the feed.

Research units

The expensive thing in the product is research: fetching public pages about a company and reading them. It is metered per company entering research.

• Adding companies and resolving identity consume no research unit. A whole CRM segment can be posted just to find out which rows Prospex can place.
• External qualification jobs consume none either. Their dispatch is bounded by the project's outstanding-job allowance, and resumes with later cycles when that is full.
• A company entering research occupies one place in the cycle budget whether Prospex or your own worker performs it. A corrected submission and an explicit retry of the same input do not occupy another.
• Prospex-executed and externally executed research are counted separately on every cycle.
• One project has at most one queued or running cycle, and the project budget is a ceiling per cycle rather than per month.

Objects

• Built lists per account follow the plan, in the table above. Imported lists do not count.
• Imported lists, projects and API credentials have no per-account cap.
• A webhook endpoint that fails 20 times in a row is switched off, and its own page says why.