Description
Runcrate, developed by Aeonmind, offers a comprehensive AI cloud infrastructure designed for AI and machine learning teams. It provides on-demand access to powerful GPUs, including H100, H200, A100, L40S, and RTX 4090, which are billed per second for most models. GPU instances can be launched and become operational within minutes, eliminating approval queues and quota requests.
Read More
The platform also features an OpenAI-compatible inference API, allowing users to call 121 different models for various tasks such as text generation, vision, image processing, video, speech, transcription, and embeddings. This unified API simplifies model integration and usage, with a single balance covering both GPU usage and model inference.
Runcrate supports a wide array of AI development activities. Users can fine-tune open models like Llama, Mistral, or Qwen with their own data using LoRA, serve open-source LLMs on dedicated GPUs with vLLM, and transcribe audio and video files at scale using the Whisper API. The platform also facilitates running ComfyUI on RTX 4090 instances and developing RAG pipelines by embedding documents and using chat models for answers.
For data persistence, Runcrate offers storage volumes that keep code, data, and checkpoints between instances, ensuring work is not lost when GPUs are terminated. Billing is precise, starting only when an instance is running and typically charged by the second, with no charges for boot time. The platform integrates with tools like Claude Code, Codex, and Cursor, allowing agents to launch and manage GPU instances directly.
Runcrate's pricing model includes per-second billing for GPU instances, per-token pricing for inference API calls, and monthly rates for storage volumes. Prepaid options are available for GPUs and inference, and users can reserve multi-node clusters for longer-term projects. The company's vision is to make AI development cheaper and more accessible by removing infrastructure bottlenecks and promoting open models.
- CategoryMachine Learning Platforms
- Pricing ModelPay-as-you-go
- DeploymentCloud
- PlatformsWeb
Runcrate's Core Features
On-demand GPU instances (H100, H200, A100, L40S, RTX 4090)
Per-second billing for most GPUs
OpenAI-compatible inference API with 121 models
Fast GPU instance deployment (typically within minutes)
Support for fine-tuning LLMs (Llama, Mistral, Qwen) with LoRA/QLoRA
Serve open-source LLMs on dedicated GPUs with vLLM
Whisper API for audio and video transcription
Storage volumes for persistent data across instances
Unified account and balance for all products (GPUs, inference, storage)
Integration with AI agents and developer tools (Claude Code, Codex, Cursor)
Prepaid options and reserved clusters available
No billing during GPU boot time
Access to Jupyter and VS Code in the browser
Ready images for fine-tuning and serving
Auto-recharge for credit threshold management
How to use Runcrate?
Connect: Add Runcrate to your preferred tool (Claude Code, Codex, Cursor, or terminal) with one command.
Sign In: Approve access through your browser, without needing to paste a key.
Launch GPU: Ask your agent to launch a specific GPU with a desired crate, e.g., "Launch an H100 with the Finetune crate."
Configure Resources: Select your GPU model and configure necessary resources.
Deploy: Launch the instance; billing starts when the GPU instance is running.
Manage: Deploy, manage, and terminate GPU instances directly from the terminal or integrated tools.
Utilize: Use the GPU for training, serving models, or call models via the OpenAI-compatible API.
Store Data: Keep your work between instances using persistent storage volumes.
Runcrate's Use Cases
- Model Training
- AI Inference
- Fine-tuning LLMs
- RAG Pipelines
- AI Agent Operations
- Image and Video Generation
- Speech and Audio Processing
- Computer Vision Tasks
- Data Processing
- Research and Experiments









