Connect a model
Bring your own weights, or point Persephone at any HTTP endpoint that speaks JSON.
Identity
How this model shows up across dashboards and reports
API connection
Persephone will call this endpoint once per test case during a run.
Must be reachable from eu-west-1.runner.persephone.io. Whitelist that IP if you have an allowlist.
Stored encrypted, never logged. Used only at request time.
POST https://api.acme-internal.co/v1/chat/completions
Authorization: Bearer sk_live_****…
Content-Type: application/json
{
"model": "model-e",
"messages": [
{ "role": "user",
"content": "{{question}}" }
],
"max_tokens": 512
}{{question}} and {{context}} are filled per test case.Test the connection
We'll send a single ping with a 64-token prompt.
Method: OpenAI-compatible · Auth: bearer · Test: not run