Limits and quotas¶
This page collects the limits that affect integrations. Where Anychat publishes no limit, this page says so explicitly.
API rate limits¶
The management API at https://api.anychat.ai/v1 does not enforce a
published per-key rate limit today. GET /v1/me carries a rateLimit
field reserved for reporting one, and it currently returns null.
Build as though a limit will arrive: retry on 429 with exponential
backoff, and read the rateLimit field rather than hard-coding a rate.
Public unauthenticated endpoints, such as redeeming an outreach link, are rate limited per client to protect them from abuse.
When a messaging provider rate-limits Anychat, the platform retries the send rather than failing it, and the send row records the outcome.
Sending limits¶
These are enforced on every send and are described in full under Sending guardrails.
| Limit | Default | Scope |
|---|---|---|
| Marketing frequency cap | One message every 48 hours | Per person, per agent, across everything that agent sends |
| Safe hours | Your configured window, any time when unset | Per agent, read in the time zone you configure; governs Automated Outreach, Campaigns and scheduled follow-ups |
| Campaign send rate | About four people a second | Per campaign |
| Automated Outreach hourly cap | You set it | Per rule |
Service-class messages and scheduled follow-ups are exempt from the frequency cap. A send you make by hand or through the API goes out when you make it: it warns rather than blocks on the frequency cap, and safe hours do not apply to it. It is refused only when your request says the person did not opt in, when the person replied STOP, or when their consent is recorded as denied.
Channel limits¶
Channels impose their own shapes, and Anychat degrades to fit rather than failing.
| Limit | Value |
|---|---|
| Carousel cards | Up to 10 on RCS, fewer on channels with a smaller maximum |
| Service window | 24 hours on RCS and WhatsApp, after the person's last inbound message |
| Early-access list | 200 numbers per channel |
| Google RBM tester invitations | 20 a day, 200 in total |
See Conversations and sessions for what the service window governs.
Content limits¶
| Limit | Value |
|---|---|
| Facts per agent | 200 |
| Facts always in view | About 40 short entries, a 3,000-token budget; beyond that the 8 most relevant ride each turn |
| Message template nesting | One level deep |
| Thing actions | Up to 4 per Thing |
Test limits¶
Tests are metered in case units, measured after each result rather than declared. Defaults:
| Limit | Default |
|---|---|
| Daily units | 250 per workspace, per UTC day |
| Units per run | 150 |
| Case samples per run | 200 |
| Units per case | 8 |
| Samples per case | 1, up to 5 |
| Turns per case | 20 as a guideline, 40 at most |
Workspace limits are set by Anychat operators. See How a Test runs.
Retention¶
| Data | Kept |
|---|---|
| Turn records | 90 days |
| Test results, including transcripts and judge output | 90 days |
| Test run summaries | 1 year |