Use case
AI companion apps
A companion app is four modalities pretending to be one person. eroq ships them behind one key: eroq RP+ or eroq RP mini for the conversation (scene or messaging register), vision so the user can send a picture, eroq Voice Turbo to speak every reply, Scribe One for call mode, Krea2 for the selfie — governed by a written acceptable-use policy that you can pass through to your users.
Opens the studio with this filled in. Sign up when you generate: 40 free credits, no card.
The request pattern
- →Chat: POST /v1/chat/completions with the persona in `context`, messaging mode for DM-style products, SSE streaming.
- →Voice: POST /v1/audio/speech with the character's voice (`char:<id>`) on eroq Voice Turbo — 2 credits per 100 characters.
- →Call mode: Scribe One transcribes the user (1 credit per minute of audio), eroq RP+ answers, Turbo speaks.
- →Pictures: Krea2 with the Character's references — the same face, 10 credits.
Why a policy wall beats a filter
The model stays in character on dark and intense turns within policy; the hard limits (no minors, no real people, nothing illegal) return an explicit content_blocked error you can handle — never a lecture mid-scene, never a charge. Deterministic enough to build moderation on.
Frequently asked questions
What does a companion conversation cost?
A reply on eroq RP mini is 1 credit, on eroq RP+ 3; a spoken reply of 200 characters adds 4 credits on eroq Voice Turbo. A hundred spoken turns on mini-plus-Turbo is about 500 credits.
Is it OpenAI-compatible?
Yes — chat completions follow the OpenAI shape; existing code ports by changing the base URL and model id.
Can my users send pictures?
Yes — image parts in a user turn, 2 credits each; the character reacts to what it sees.
Keep exploring