Connect the models that write and illustrate Publisher articles while keeping keys, model access and costs under your control.
The text-provider connection test validates text generation only. OpenAI or Gemini image generation must be verified through an image-enabled draft job.
Text-provider fields
| Field | Default or values | Purpose |
|---|---|---|
| Provider | OpenAI; Gemini; Anthropic; xAI. Default: OpenAI | Selects the API family used for research, planning and writing |
| OpenAI model | Default gpt-4o-mini |
Text model used when OpenAI is selected |
| Gemini model | Default gemini-2.5-flash |
Text model used when Gemini is selected |
| Anthropic model | Default claude-3-5-haiku-latest |
Text model used when Anthropic is selected |
| xAI model | Default grok-2-latest |
Text model used when xAI is selected |
| Custom model ID | Provider-supported identifier | Overrides presets when you need a compatible model not listed in the UI |
| Creativity | Default 0.72 | Balances deterministic adherence and variation; extreme values can reduce consistency |
Image-provider fields
| Field | Default or values | Purpose |
|---|---|---|
| Image provider | OpenAI or Gemini; default OpenAI | Selects the image API |
| Image model | Default gpt-image-1-mini for OpenAI |
Model used to create featured and inline images |
| AI image quality | Default Medium | Controls provider-generation quality, not WebP compression |
| API key | Customer-owned secret for the selected provider | Authorises image requests; may differ from the text provider |
Save secrets safely
Publisher encrypts stored secrets with AES-256-CBC when OpenSSL is available and uses a compatibility fallback otherwise. A blank secret field preserves the existing saved value. Never paste keys into screenshots, support messages or article content.
Recommended setup order
- Choose one text provider and confirm that the account has billing and access to the selected model.
- Paste the matching key, select a preset model and save.
- Run the connection test. Resolve authentication or model errors before continuing.
- Create one short, text-only Draft.
- Choose an image provider, model and Medium quality; enable one inline image.
- Create a second Draft and confirm that the Media Library receives the image and alt text.
Choosing models
Use a fast, economical model for routine high-volume work and evaluate a stronger model on difficult technical or brand-sensitive topics. Compare complete article reports rather than isolated paragraphs. A custom model ID is useful only when the provider account actually exposes that model and its API remains compatible.
Automatic behaviour
Publisher sends structured requests for research, brief, section plan and writing. Temporary retryable failures can be requeued with backoff and resumed from a checkpoint. It does not purchase provider access, select billing limits or guarantee that a provider keeps a model available.
Common errors
- 401/unauthorised: wrong, revoked or mismatched provider key.
- Model not found: the saved model ID is unavailable to that account or region.
- Quota/rate limit: inspect provider billing and limits; allow retry backoff instead of repeatedly starting jobs.
- Text works, images fail: verify the image provider, model, permissions and image billing separately.
- Output ignores instructions: first inspect the Content settings and conflicting additional instructions; changing providers may not solve a configuration conflict.
Verification checklist
- The selected provider and saved key belong together.
- The text connection test passes.
- A Draft completes with the chosen text model.
- An image-enabled Draft stores WebP media with alt text.
- No complete key appears in logs or screenshots.