Settings
Default AI Provider
Pro users can set the default AI provider and model used throughout the app, such as when creating new projects or using AI-powered features. The chosen provider and model combination will be applied as the default across all AI interactions.
Supported Providers
OpenAI
Cloud-hosted LLM provider
Anthropic
Cloud-hosted LLM provider
Google Gemini
Cloud-hosted LLM provider
DeepSeek
Cloud-hosted LLM provider
xAI (Grok)
Cloud-hosted LLM provider
Ollama
Run models locally on your machine
Per-Provider Configuration
Default Model
Select the default model for each provider (e.g., GPT-4o, Claude Sonnet, Gemini Pro).
Temperature
Adjust the creativity/randomness of responses per provider.
Max Tokens
Set the maximum token limit for generated responses.
Processing Mode
Choose between Standard, Batch, or Realtime processing modes per provider.
Ollama (Local)
Ollama runs models locally on your machine. Additional options include local server status monitoring, model installation and removal, and a cloud acceleration mode for offloading processing when local hardware is limited.
API Keys
Managing API Keys
Enter and manage API keys for each AI provider. Keys are stored locally on your machine and are never sent to Kaiju Hub AI servers. Each key entry includes a show/hide visibility toggle, a clear button, key prefix validation to catch formatting errors, a link to the provider's documentation, and a connection status indicator.
Supported Keys
OpenAI
API key for OpenAI models. Key prefix validation ensures correct format.
Anthropic (Claude)
API key for Anthropic Claude models.
Google Gemini
API key for Google Gemini models.
DeepSeek
API key for DeepSeek models.
xAI (Grok)
API key for xAI Grok models.
Ollama
Optional API key for Ollama. Includes a server status indicator and refresh button to verify your local Ollama instance is running.
HuggingFace
Token for accessing gated models used in image generation and local model downloads.
Agentic Agents
Overview
Configure autonomous coding agents that work within your project workspace. Each agent supports CLI, SDK, or ACP operating modes. Set default permissions, configure max turns, and manage which agents are available for new projects.
Agent Types
Claude Code
Anthropic's autonomous coding agent. Supports CLI/SDK/ACP modes. Configure max turns for extended sessions.
Codex
OpenAI's coding agent. Supports CLI/SDK/ACP modes.
Gemini CLI
Google's coding agent. Supports CLI/SDK/ACP modes.
Custom Agents
Add third-party agents with custom command configuration (e.g., OpenCode). Define the executable command and arguments for each custom agent. Supports both Internal and External modes based on the toggle setting.
ACP Agents
Install agents from the ACP (Agent Communication Protocol) registry. Includes install detection to verify agent availability on your system.
Configuration Options
Default Permissions
Set the default permission mode for new projects: "Default" (prompt for each action) or "Skip All Permissions" (fully autonomous without confirmation prompts).
Max Turns (Claude Code)
Configure the maximum number of turns Claude Code can take before pausing for user input.
Default Modes
Set the default agent mode (CLI, SDK, or ACP) for new projects.
Default External Terminal
Choose the default external terminal for agent operations (e.g., PowerShell, Command Prompt, Windows Terminal).