On-device 3B LLM for agent planning

Octo-planner Highlights

Edge Optimized

On-device planning AI planner model based on Phi-3 Mini, a 3.8 billion-parameter language model optimized for edge devices.

High Success Rate

Highly accurate achieves a 97% success rate in in-domain test environments after fine-tuning.

Modular Execution

Planner-Action framework Separates planning and action execution into distinct components, improving modularity and specialization.


Multi-domain capability employs a novel multi-LoRA training method to handle complex, multi-domain queries while maintaining efficiency.

Octo-planner Demo Video

On-device planning model for AI agents

Octo-planner, an innovative open-source planning model based on Phi-3 Mini with 3.8 billion parameters, represents Nexa AI's advancement in applying large language models (LLMs) for on-device planning. Octo-planner introduces a novel Planner-Action framework that separates planning and action execution. This approach allows it to achieve high accuracy comparable to larger models while significantly improving efficiency for edge devices. Octo-planner also employs a multi-LoRA training method to handle complex, multi-domain queries without compromising performance.

Explore our collection of 200+ Premium Webflow Templates