Rate limits
Two sets of limits apply to what your AI client does through FloConnector: FloConnector's own fair-use pacing, and the vendor's real API quotas behind it. FloConnector's job is to keep you inside both without you having to think about either.
Tool-call limits
FloConnector paces tool calls on two rolling windows. A call needs room in both to go through.
| Window | Default limit | Scope |
|---|---|---|
| Per minute | 60 calls / min | Per profile |
| Per day | 20,000 calls / day | Per workspace |
The per-minute rate is the one you would ever notice. The daily ceiling sits well above normal use and only exists to stop a runaway client. Both are rolling budgets, so short bursts are fine as long as the average stays under.
What is not counted. Connecting, signing in, the initial handshake, and listing the available tools are all free. Only actual tool calls, the ones that do work, are paced.
The per-minute rate is scoped to the profile, the endpoint you hand to a client, so each connected client gets its own budget and two people on different profiles never compete. The daily ceiling is scoped to the whole workspace, as a tenant-wide backstop across every profile in it.
What happens when you hit one
Nothing breaks. The tool returns a normal result telling the model it was paced and how many seconds to wait, so the client simply pauses and tries again. A paced call is not a billable call, so staying inside the limit never costs you anything extra.
Vendor quotas behind FloConnector
Each vendor API has its own quotas on top of FloConnector's. FloConnector's adapters handle pagination, back-off, and retry so a single tool call stays within them. When a vendor is genuinely throttling, the tool surfaces it in its result with a retry hint rather than hammering the API.
Composite tools help here too: folding many vendor calls into one is the most effective way to stay well under any quota. See How FloConnector works.
Signing in and the dashboard
A few control endpoints, the sign-in code requests and the public connector
directory, are rate-limited per IP to keep them from being abused. If you somehow
trip one, for example by requesting many sign-in codes in a row, you get a short
wait and a Retry-After hint. Normal use never comes close.
A sign-in code carries its own separate limit: five wrong attempts and that code is retired, so request a fresh one rather than guessing again.