• Autonomous Reasoning: Employs a ReAct (Reasoning and Action) loop to solve complex tasks using integrated tools.
• Intelligent Model Routing: Automatically classifies tasks to utilize the most efficient model, ranging from lightweight 8B models to flagship 120B+ architectures.
• Native Vision Support: Seamlessly processes local image files (PNG, JPG, WebP) using state-of-the-art multimodal vision models.
• Resilient API Client: Features automatic key rotation and sophisticated error handling for high-availability performance.
• Secure Configuration: Implements OS-native secure storage for API credentials with restricted file permissions.
• Premium Terminal UI: Minimalist, branded interface featuring real-time status indicators and syntax-highlighted output.

POLARIS-CLI provides the agent with a suite of low-level system capabilities:

• Filesystem: read_file, write_file, and ls_dir for direct file manipulation.
• System: run_cmd for terminal execution and sys_get_info for environment awareness.
• Search: search_code for fast, recursive grep-style regex searching across the codebase.

The agent dynamically routes tasks to specialized models based on complexity and intent:

• Flagship Logic: openai/gpt-oss-120b (System Architecture)
• Deep Reasoning: qwen/qwen3-32b (Complex Coding & Math)
• Heavy Generation: llama-3.3-70b-versatile (Long-form content)
• Native Vision: meta-llama/llama-4-scout-17b-16e-instruct (Multimodal analysis)
• Light Execution: llama-3.1-8b-instant (Greetings & Terminal commands)