Model Provider

Use Vidjutsu as the model boundary. Do not ask the user for upstream model, Gateway, or runner credentials.

Authenticate

The production API base is https://api.vidjutsu.ai. Raw API calls use:

Authorization: Bearer <VIDJUTSU_API_KEY>

For CLI workflows, store the key once:

vidjutsu auth --key "$VIDJUTSU_API_KEY"
vidjutsu usage

An active $99/month subscription is required for metered intelligence and generation operations. Vidjutsu enforces daily request limits; it does not charge per-operation fees.

Execution ownership

• Clone-check, character creation, and starting-image generation run through

the authenticated tenant's scoped Vercel AI Gateway allocation.

• Clone video is Kling 3.0 Motion Control only. Vidjutsu dispatches it to a

dedicated runner and receives authenticated lifecycle callbacks.

• The caller stores only the Vidjutsu clone task ID and reads status through

GET /v1/clones/video/{id}. Never call or poll a provider directly.

• No alternate clone-video model or fallback is supported.

External services

ZapCap captions, Zernio publishing, native platform analytics, and optional user-supplied music are separate integrations because Vidjutsu currently has no equivalent endpoints. Keep their credentials and billing separate; never describe them as tenant Gateway operations.

Key behaviors

• Default to model: "kling"; no other clone model is accepted.
• Check vidjutsu usage before a multi-stage workflow.
• Preserve accepted task IDs across retries and resumes.
• Never expose or attempt to obtain the tenant's underlying Gateway key.