Model reference · open weights

Moxie

Available as managed deployment Licence fee LLMs turtle89431 · community Vision + text 1 variants 1k dl/mo

Moxie is an open-weight language model from turtle89431. AxForge deploys and operates it for you on dedicated EU-owned hardware — with the licence handled where one is required.

Available as managed deployment — configured and operated for you on dedicated EU hardware, quoted per deployment.

What it is

Released byturtle89431
TypeLanguage models
TaskVision + text
Based onturtle89431/Moxie
Released2026-09-09
Popularity1k downloads / month
LicenceCommercial licence needed

About

What Moxie is

Model Card: Moxie Developer: Moxiegen Business Group Deployment: Moxie Cloud service (via Moxie Desktop application)

1. Model Summary Moxie is a specialized Large Language Model (LLM) and agent that serves as a demonstration of the "Moxiegen Method"—a proprietary algorithmic framework designed to strip redundant weight and computational overhead from LLMs. The end result of applying the Moxiegen Method is the ability to perform native inference with minimal GPU requirements, making previously impossible-to-run models accessible on consumer-grade hardware. Moxie demonstrates this capability by offering a suite of open weight models that have been algorithmically optimized by the Moxiegen method.

The foundation of the Method is a Lossless Pointer-Based Weight Mapping algorithm. Instead of storing redundant floating-point values repeatedly, the algorithm scans the model and maps identical numerical values to a single, centralized pointer reference. This process is entirely lossless: mathematical computation remains in full FP32 precision, but the memory overhead of storing duplicate weights is eliminated, allowing a 235-billion-parameter MoE model to run at speeds exceeding 160 tokens per second on a sub-$1,000 refurbished workstation.

Read the full model card

Moxie is accessible through the Moxie Desktop application that is downloadable at Moxiegen.com. The application allows users to access the Moxie Cloud service, and to initiate a download of the Moxie LLM for native inference. The Moxie LLM is available as light, standard, and full versions, offering a range of inference capabilities.

2. Technical Specifications

  • Architecture: Optimized transformer-based model.
  • Optimization: Custom weight-pruning/redundancy-removal techniques.
  • Performance Profile: High-speed, low-latency, optimized for "small-but-mighty" or "fast-and-efficient" use cases.
  • Infrastructure: Hybrid model hosted on the Moxie Cloud service, enabling complex computations (like image generation and heavy file processing) to be offloaded from the local machine while maintaining a seamless desktop experience. Full local hosting can be enabled in the user interface.

3. Model Lineage & Base Architectures While Moxie utilizes custom weight-reduction techniques, its underlying reasoning and instruction-following capabilities are built upon a distillation of leading open-weight models:

  • Qwen-based foundations: Utilized for superior coding and mathematical reasoning logic.
  • GLM-based foundations: Leveraged for efficient long-context window management and instruction adherence.
  • Gemma-based foundations: Integrated for lightweight, high-speed prompt-to-action processing (ideal for desktop automation tasks).

Note: The Moxiegen Method prunes the overlaps in these weights to create a specialized, lighter, and faster-running version.

4. Technical Stack (Libraries) Moxie’s execution environment and the way it processes files and certain logic-heavy tasks rely on the following standard and specialized libraries:

  • torch / transformers: For managing the underlying tensor-based weight structures and model inference.
  • numpy: For high-speed numerical processing during data manipulation.
  • pandas: For heavy-duty processing of structured data (CSV, Excel, etc.).
  • Pillow (PIL): For image-based manipulation and processing (when handling local or cloud-fetched images).
  • scikit-learn: For certain internal classification/clustering tasks used in sub-agent task decomposition.

5. Data, Storage, and Infrastructure Primary Weights Storage (Buckets):

  • moxie-model-weights-v1: Stores the pruned, optimized weight-sets for all active model versions.
  • moxie-cloud-cache: Temporary storage for active session context and intermediate computation results.
  • Training/Fine-tuning Datasets:
  • moxie-instruction-set: A proprietary dataset of high-quality, tool-use-focused instructions.
  • moxie-desktop-automation-v2: A specialized dataset focused on system-level actions (UI, File, and Shell commands).
  • Data Integrity: All model weights are verified via MD5 and SHA-256 checksums during deployment to ensure no corruption during the weight-reduction process.

6. Licensing & Compliance

  • Model Weights: Proprietary (Moxiegen-owned, optimized via custom pruning).
  • Inference Engine: Proprietary (Moxie Cloud service).
  • Upstream/Base Models: Licensed under their respective open-weight licenses (e.g., Apache 2.0 or Gemma-specific licenses).
  • Usage Policy: Intended for use within the Moxie Desktop application, as licensed by the privacy agreement and user agreement located at Moxiegen.com.

7. Intended Use Moxie is intended to serve as a general inference engine and AI assistance for non-commercial desktop users. Primary use cases include:

  • Desktop Automation: Executing shell commands, managing files, and controlling system processes.
  • Data Analysis: Reading, writing, and modifying complex file formats (Excel, Word, SQLite, CSV, etc.).
  • Workflow Orchestration: Managing complex, multi-step tasks using a planning-and-execution (Plan/Execute) or sub-agent architecture.
  • Content Creation: Generating or analyzing images, video, music, and text-based content.
  • Information Retrieval: Web searching, web scraping, and processing large-scale document data.

8. Capabilities & Toolset Moxie has direct access to a suite of specialized tools, including:

  • System Control: Manipulation of windows, mouse, keyboard, and running processes.
  • File System Operations: Full CRUD (Create, Read, Update, Delete) permissions over a wide range of file types.
  • Web & Data: HTTP requests, web searching, and scraping.
  • Advanced Computation: Media generation, and heavy mathematical or script-based execution.
  • Memory Management: Persistent, long-term memory storage for user preferences and project-specific facts.

9. Limitations

  • Cloud Dependency: While the app is a desktop application, the primary intelligence resides in the Moxie Cloud service. Perform

From the published model card. Full card on the HuggingFace links in the sidebar.

How it works

How language models work

Your prompttext / messagesTransformerattention over tokensNext-token loopgenerate + streamResponsetext · tool callsA language model reads your tokens and predicts the next one, again and again, streaming the reply back.

Using it via the API

Call it like any OpenAI endpoint

Once AxForge deploys moxie for you, it answers on the OpenAI-compatible API — the same base URL and keys as every other model. (moxie below is illustrative; you get the exact model name on deployment.)

$ curl -sS https://api.axforge.ai/v1/chat/completions \
  -H "Authorization: Bearer $AXFORGE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"moxie","messages":[{"role":"user","content":"Hello"}]}'

Create an account — your API key is available in the console. 3M free tokens every 30 days with every new account.

© 2026 AxForge · EU-hosted AI infrastructure Pricing Docs Trust Privacy Terms