Configuration & Providers — KuroCodeAgent 2.0 Docs

# Supported LLM Providers

KURO 2.0 uses a task-aware model router with automatic fallback cascade. Supported providers include:

1. Google Gemini

Default primary provider for reasoning, vision, and tool orchestration.

2. Groq

Ultra-low latency execution via Llama-3.3-70b-versatile.

3. OpenRouter

Unified gateway for Claude 3.5 Sonnet, DeepSeek, and open models.

4. OpenAI

GPT-4o and standard OpenAI compatible endpoints.

5. Local Ollama (Offline / Air-Gapped)

Runs completely locally on your machine with zero external network transmission.

# Encrypted DPAPI Vault Configuration

Rather than leaving sensitive API credentials in plain text files on disk, KURO provides a built-in cryptographic vault:

terminal (inside kuro)

/key gemini AIzaSyYourApiKeyPlaceholder
/key groq gsk_YourGroqKeyPlaceholder
/vault status

* Security guarantee: On Windows, keys are encrypted using the current user's OS credentials via CryptProtectData. On Linux/macOS, keys are sealed using PBKDF2-HMAC-SHA256 with 200,000 iterations.

# Environment Variables Reference

For headless CI/CD runners or server environments, you can configure options via a .env file:

.env file example

# Primary LLM Provider Keys (Use your own credentials)
GEMINI_API_KEY=your_gemini_key_here
GROQ_API_KEY=your_groq_key_here
OPENROUTER_API_KEY=your_openrouter_key_here
OPENAI_API_KEY=your_openai_key_here

# Runtime Orchestration Tuning
KURO_PROVIDER=gemini
KURO_MAX_STEPS=10
KURO_SAFE_MODE=true
KURO_GIT_CHECKPOINTS=true

Variable	Default	Description
KURO_PROVIDER	gemini	Specifies initial primary provider. Fallback triggers on 429 or network failure.
KURO_MAX_STEPS	10	Maximum sequential problem-solving steps before demanding confirmation.
KURO_SAFE_MODE	true	Enforces permission prompts for write operations and command execution.
KURO_GIT_CHECKPOINTS	true	Automatically snapshots git repository before modifying files for /undo.