Limits
Requests per key, calls at once, test keys, per-address limits, body sizes and how long a call may run.
Limits keep one integration from crowding out another, and keep a call's cost bounded. When you reach one, the answer says which, and when to try again.
Requests per key
600 requests a minute per secret key. The minute is a fixed window that starts with the key's first request. Past it, the answer is 429 rate_limited with Retry-After: 60. No other rate-limit headers are sent.
- Requests with a bad key aren't counted.
- A key and its rolled-over old secret share one budget.
- The MCP server counts against the same budget: every MCP request, and every
/v1request a tool makes for you.
Calls at once
10 calls at once per organization, unless we've raised it for you. Every key, the website widget and the dashboard count together. The next start is 429 too_many_calls with Retry-After: 30.
A meeting counts once it has been sent on its way; a meeting scheduled for later doesn't count until then.
Test keys
A test key's calls are never charged, and so they are capped:
- 60 seconds at most, for a call or a meeting.
- 20 calls per organization in any 24 hours, then 429
test_limit. It has noRetry-After: the count is over the last 24 hours.
Per address
In front of the api, each address may make about 10 requests a second (in bursts of up to 40), with at most 100 open at once. That leaves room for a call's sound and live events sockets for every call your organization can run. Past it, the proxy answers a plain-text 429.
Body size
Our public addresses (api.daxov.com, app.daxov.com) passes a request body of up to 11 MB; a larger one is refused before it reaches the api, with a plain-text 413. Within that:
| What | At most | Past it |
|---|---|---|
| A JSON body | 2 MB (10 MB on POST /v1/documents) | 400 invalid_json |
| A document file (knowledge base) | 10 MB | 413 file_too_large |
| A launcher icon (icons) | 64 KB | 413 file_too_large |
| A cloned voice's recordings (cloned voices) | 10 MB each, and 11 MB for the whole request through our public address | 413: file_too_large from the api, or the plain-text one past 11 MB |
| A request to the MCP server | 4 MB | 413 |
The api itself would take 25 MB of recordings for one clone, but through api.daxov.com a request over 11 MB never reaches it: keep one clone's recordings under 11 MB together.
How long a call runs
A call's max_seconds is the least of:
- what you ask for in
max_seconds; - the character's
max_call_seconds(30 to 1,800 seconds; 300 by default); - 60 seconds, for a test key;
- what your organization's plan minutes and balance cover.
On a paid organization, a call that would get less than 30 seconds is refused with 402 insufficient_balance, so don't ask for less than 30 there. The call ends at max_seconds: the balance can't run out mid-call. The browser must connect within 120 seconds of the start.
A meeting's max_seconds is 60 to 10,800 (3 hours), 3,600 by default; the character's max_call_seconds doesn't apply to it. See Meetings.
What an organization can keep
| What | At most |
|---|---|
| Webhook endpoints | 10, live and test together |
| Documents | 200; a call's knowledge, the character's own and its documents' together, 120,000 characters |
| Guardrail sets | 100, and 10 per character |
| Pronunciation dictionaries | 50, and 5 per character |
| Tools | 50, and 20 per character |
| MCP servers | 10, and 5 per character |
| Cloned voices | 5 |
Past one of these, the answer is 409 (webhook_limit, asset_limit, voice_limit or knowledge_too_large).