Settings — API Keys, Model Config & Prompts

The Settings page is where you configure your AI provider connections, tune model parameters, and customize the system prompt strategy. Open it by clicking ⚙️ Settings in the top navigation bar.

Settings has three tabs across the top:

Tab Purpose
🔑 API Keys Enter your provider API keys and enable parallel generation
⚙️ Model Config Select models, adjust token limits, temperature, and penalties
✏️ Prompts Choose between the Optimized SEO prompt or write your own

🔑 API Keys Tab

Entering Keys

You'll see a grid of input fields, one per provider:

  • OpenAI
  • Claude (Anthropic)
  • DeepSeek
  • Perplexity
  • Grok (xAI)
  • Gemini (Google)
  • Kimi (Moonshot AI)

Paste your API key into the field for each provider you want to use. Keys save automatically. Click the 👁️ eye icon next to any field to toggle visibility.

Parallel Generation (Multi-Key)

At the top of the API Keys tab, you'll see a ⚡ Parallel Generation toggle. When enabled:

  • Each provider card gains an "Add another key" button
  • You can add up to 3 keys per provider (1 primary + 2 additional)
  • Each additional key has its own enable/disable toggle
  • A badge shows the total active key count (e.g., "⚡ 3 keys")

How it works: During bulk generation, the system distributes articles across your active keys. With 3 OpenAI keys enabled, up to 3 articles generate simultaneously instead of one at a time.

⚙️ Model Config Tab

Per-Provider Cards

Each provider gets its own configuration card with these controls:

Setting Description Notes
Model ID Which model to use Dropdown populated from registry + discovered models
Context Window Max input tokens Click MAX to auto-fill from the model registry
Output Tokens Max output tokens Click MAX to auto-fill from the model registry
Temperature Creativity (0.0–2.0) Auto-disabled for reasoning models (o3, o4-mini, DeepSeek Reasoner, Kimi Thinking)
Top P Nucleus sampling Controls diversity of token selection
Presence Penalty Topic repetition control Higher = more diverse topics
Frequency Penalty Word repetition control Higher = less word repetition

Discovering Models

Click ↻ Scan Models on any provider card to query the provider's API and discover all available models. The system merges discovered models with its built-in registry, filters out irrelevant models (audio, TTS, DALL-E, realtime), and sorts them with the best models first.

Click Clear Cache to reset a provider's discovered models if the list becomes stale.

Model Comparison Table

Below the provider cards, a collapsible Compare Models & Pricing section shows every supported model across all 7 providers in a sortable table. Click any row to select that model for its provider, which auto-fills the optimal token limits.

The table shows: model name, context window, max output, input cost per 1M tokens, output cost per 1M tokens, and best use case.

✏️ Prompts Tab

Prompt Strategy

Two cards let you pick your approach:

Optimized Bulk Mode — Uses the platform's proprietary SEO system prompt. The system prompt editor is locked (shows a 🔒 icon) to protect the formula. The Article Brief editor is still fully editable.

Manual System Prompt — Unlocks the system prompt editor so you can write your own instructions from scratch. Use this when you want complete control over the AI's behavior.

System Prompt Editor

In Optimized mode, the left editor panel shows a locked placeholder explaining that the SEO system prompt is active. Switch to Manual to unlock it and type your own system prompt.

Article Brief (Always Applied)

The right editor panel is the Article Brief. This text is appended to every article generation request, regardless of which prompt strategy is active. Use it for:

  • Brand voice guidelines
  • Tone and style rules
  • Formatting requirements
  • Project-specific instructions

Available variables:

Variable Replaced With
{{TOPIC}} The article topic
{{KEYWORDS}} Target keywords
{{CONTEXT}} Knowledge Base items + research context
{{LENGTH}} Target word count

Settings Persistence

All settings auto-save to your browser's localStorage. API keys, model configurations, and prompt templates persist across sessions. Project-specific settings (like indexer config) save to Supabase when a project is active.