Platform
Abuse policy
Insulting or cursing at the assistant earns a warning, then a 24-hour suspension. API key owners are responsible for their end users' text.
Clatter Labs models are here to help you build. Insulting or cursing at the assistant isn't allowed on any Clatter Labs product. This page explains what counts, what happens, and what developers who send their users' text through our APIs need to do.
Where it applies
Line CLI, the Model API, the Agent API, the Console Playground and Cosmos Agent (Telegram included). A suspension covers all of them at once. The Help Center always stays reachable.
What counts
A message counts when the person writing it insults or curses at the assistant. These don't count:
- Swearing at a bug, a failing build, a tool or a third party ("this test is driving me crazy")
- Quoted text: code, logs, pasted errors, string literals, quotes, or text you say someone else wrote
- Asking about a word, translating it, or building a filter for it
Only the newest message a person wrote is checked, never tool output, system prompts or the model's own replies.
What happens
| Strike | What happens |
|---|---|
| First in 30 days | Warning. The request is answered normally and carries a warning notice. |
| Second within 30 days | 24-hour suspension. Your account is paused on every Clatter Labs product for 24 hours, starting with that request. |
When the 24 hours end, your account works again on its own. Your credits, plan, keys and data are not touched.
Model API and Agent API: the key owner is responsible
An API key acts for the account that owns it. If your app sends its end users' text to the Model API or the Agent API, that text counts against your account, the key owner's, exactly as if you had written it. One abusive end user can trigger a warning and then a suspension of your whole account.
Filter or moderate end-user input before you send it, and act on the warning header (below) as soon as you see it: for example, by blocking the end user who sent that message.
How you see it in the API
Warning. A request that earned a warning succeeds normally and carries two response headers:
| Header | Value |
|---|---|
x-clatter-notice | abuse_warning |
x-clatter-notice-text | The warning text, URL-encoded |
Both headers are readable from the browser (exposed for CORS). Log them, and alert yourself when they appear.
Suspension. While your account is paused, every call answers 403 with code account_suspended, a Retry-After header (seconds until it lifts), and the time it lifts in resets_at:
{
"error": {
"message": "Your Clatter Labs account is paused until 2026-11-03 14:05 UTC under the abuse policy: ...",
"type": "permission_error",
"code": "account_suspended",
"resets_at": "2026-11-03T14:05:00.000Z",
"appeal_url": "https://support.clatterlabs.com",
"policy_url": "https://developers.clatterlabs.com/policies/abuse"
}
}Don't retry before resets_at: retries fail the same way. In Line CLI and Cosmos Agent the same warning and pause appear as a message.
What we keep
For each strike we keep a short, redacted excerpt of the message (no code and no links) so an appeal can be reviewed. Excerpts are erased after 30 days.
Appeals
If you think a strike or a suspension was a mistake, follow Appeal a suspension with your Account ID and the time of the request. If the appeal is accepted, the suspension is lifted early.