Configure model services
Choose how to pay for AI services, configure the connection and verify it without generating media.
Do you need a model service?
Trimming, arranging existing footage, editing existing captions and local export do not require a generation model key. Video analysis, speech transcription and generating new images, video or audio require the corresponding model service. Configure only the services your task needs.
1. Choose your billing route
With your own provider account, the provider bills that account. With an official rikoko connection, use only the service address and personal API key actually issued to you; official availability and credits must be checked on your account. Website login, a rikoko personal API key, a provider key and a local workspace key serve different purposes. Do not use them interchangeably.
The web app uses the cloud account and services configured by its operator. Changing your local MCP files does not configure the web app. A model shown in either interface is not a guarantee that it is enabled for your account.
2. Locate the active configuration
For a managed installation, find the installation.json file referenced by VIDEOWORKBENCH_MANAGED_CONFIG in the client’s rikoko MCP entry. Its home field identifies the installation directory. The files below are relative to that directory. Source or self-hosted deployments may use a different configuration path; inspect their launch configuration instead.
| File | What belongs here |
|---|---|
| config/models.json | Connection mode, provider adapters, model profiles and input/cost limits. No secret key values. |
| config/credentials.json | Private key values, indexed by the exact keyEnv name in models.json. Enter secrets locally, not in chat. |
| data/server.log | Startup and connection diagnostics. Remove secrets before sharing excerpts. |
3. Configure a compatible profile
For your own provider account, copy this request and tell the agent which provider you use. It can locate the files and prepare the configuration with you. For an official connection, ask it to use your issued service address instead. Manual configuration details follow the request.
Help me configure AI services for my existing rikoko installation. First identify its version, active configuration files and billing route without displaying secrets. I want to use my own provider for video generation, preferably a compatible MiniMax profile. Check the release profile, reference support and current price before proposing changes. Preserve other connections. Tell me where to enter the key privately. Do not enable paid execution or submit a task until I approve the configuration and budget.For your own provider, models.json uses schemaVersion: 1, mode: "custom" and a providers array. Each connection has id, adapter, keyEnv, enabled and models. Supported adapter identifiers are fal, fal-video, replicate and gemini. Each model profile needs an exact endpoint, kind, parameters, priceVersion, maxCostCents and hardBoundConfirmed; reference models also need the correct input bindings. A model brand or API key alone is not enough.
Have your agent inspect a compatible profile from your installed release and verify its endpoint, input limits and current provider price. Do not guess costs or mark hardBoundConfirmed as true before checking the cost bound. Keep enabled false while preparing the configuration. Store the credential privately under the matching keyEnv name; on macOS/Linux the credentials file must remain owner-only (0600). Back up existing configuration and preserve other connections.
For an issued official connection, use mode: "official" with official.url set to the supplied service origin and official.keyEnv set to the credential name. Do not invent a cloud URL or reuse a provider key. If no official service address and key have been issued, use your own supported provider or continue with local editing.
4. Apply and verify
After reviewing the profile and spending limit, enable the chosen connection. Finish active jobs, restart the local service using your installation’s management procedure, then reconnect the client. Merely reloading a chat does not reload a running backend’s configuration. Ask for a read-only check of capabilities and generation_usage.
Check rikoko capabilities and generation_usage. Report the billing source, exact model IDs, availability, supported reference inputs and output settings. Explain any missing configuration. Do not analyze media or generate anything.The check should identify the expected model and billing route. It does not test provider balance or actual output quality. If a model is unavailable, check enabled, the matching credential name, profile validation and cost confirmation. If an old catalog remains, check the active installation and restart. A first paid test is a separate step with its own agreed budget.