cagataydev/loopl-2b

VISIONPricing:Input $0.32 / Cached $0.064 / Output $1.6Concurrent Unit Cost:1Model Size:2.3BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Oct 6, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

The cagataydev/loopl-2b is a 2.3 billion parameter vision-capable agent model, post-tuned from Qwen3.5, designed for on-device execution. Developed by Cagatay Cali, it integrates an agent loop and tools directly into its weights, enabling robust tool calling and identity retention without a system prompt. This model excels at running agents on mobile devices, particularly for tasks involving phone interaction, image processing, and user-specific training, even offline.

Loading preview...

loopl 2B · v1.2: On-Device Agent Model

The cagataydev/loopl-2b is a 2.3 billion parameter model, post-tuned from Qwen3.5, specifically engineered to function as an intelligent agent directly on mobile devices. It integrates an agent loop, identity, and tool-calling manners directly into its weights, reducing reliance on explicit system prompts. This model is part of the larger loopl ecosystem, which includes a Swift SDK and a free iPhone app, enabling agents to interact with phone functionalities, process images, and operate offline.

Key Capabilities

  • On-Device Agent Execution: Designed for efficient operation on mobile devices, including offline capabilities.
  • Integrated Tool Calling: Supports native and loopl JSON tool dialects, allowing the agent to interact with phone features like torch, haptics, translation, and other apps.
  • Vision Capabilities: Retains the base Qwen3.5's vision tower, enabling on-device image reading and processing.
  • Robust Identity: Maintains a consistent identity and behavior, even without a system prompt, due to fine-tuning.
  • Trainable: The bf16 weights serve as a base for continual post-tuning, allowing users to train the model on their own conversations directly from the app.
  • Swift SDK Integration: Seamlessly integrates with the loopl Swift SDK for building custom on-device agents.

Good For

  • Mobile AI Applications: Ideal for developers building intelligent agents that need to run locally on smartphones.
  • Offline Functionality: Use cases requiring AI capabilities without continuous network access.
  • Tool-Augmented Agents: Applications where the agent needs to interact with device hardware or other apps.
  • Personalized AI: Scenarios where users want to fine-tune the agent's behavior with their own data.
  • Vision-Enabled On-Device Tasks: Tasks involving image analysis or interaction directly on the device.

Finetunes
2 models