FlexAI
Unleash AI power universally, efficiently, and sustainably.. [Contact for Pricing]
Last verified:
What is FlexAI?
FlexAI is an AI infrastructure platform that orchestrates inference, fine-tuning, and training workloads across any cloud or hardware provider. It enables AI builders, startups, growing AI teams, and enterprises to deploy production AI without managing infrastructure themselves. The platform handles hardware, software, networking, and storage layers, allowing teams to deploy once and let FlexAI handle the rest.
Key features include TokenFactory for getting tokens in seconds, OpenAI-compatible API supporting any model, auto-scaling for cost and performance optimization, managed checkpoints for training jobs, built-in observability with hardware metrics dashboards and experiment tracking, and a CLI plus Web Console for workload management. The platform supports Nvidia GPUs (50,000+ deployed), offers 67% average cost savings, and achieves under 24 hours time to first production deploy.
FlexAI is designed for builders creating AI solutions, startups needing fast deployment, growing AI teams managing compute across multiple clouds, and enterprises requiring secure, governed infrastructure with RBAC, multi-tenancy, and VPC/on-prem/air-gapped deployment options through their CloudFoundry offering for system integrators.
FlexAI pricing
Pricing model: Freemium
New accounts come with $5 free to get started. The platform offers cost optimization where you pay only for what you use. Average cost savings of 67% compared to traditional infrastructure. For enterprise pricing, custom plans, and credit availability, users can book a demo or talk to the FlexAI team. Cloud Services include inference, fine-tuning, and training on any cloud with one managed platform.
FlexAI pros
- 67% average cost savings on AI infrastructure
- 50,000+ GPUs deployed across clouds
- Deploy once, FlexAI handles infrastructure automatically
- Under 24 hours time to first production deploy
- OpenAI-compatible API works with any model
- Auto-scaling optimizes cost and performance
- No DevOps overhead required
- Managed checkpoints with automatic resume capability
- Built-in observability with hardware metrics dashboards
- Works across any cloud provider or hardware
- Tokens in seconds with TokenFactory
- Secure secrets management without committing to codebase
- Support for single GPU to multi-node training
- Python 3.11, PyTorch 2.4, CUDA 12.4 pre-configured
- Web Console and CLI for flexible workload management
- Dataset manager supports any file type with no size limit
- Real-time training progress monitoring with TensorBoard
- 30-day data retention for infrastructure metrics
FlexAI cons
- CLI-based interface only (web GUI coming soon)
- Managed checkpoints limited to 5GB maximum size
- Python 3.11, PyTorch 2.4, CUDA 12.4 stack fixed
- Only Nvidia GPU hardware supported
- Model must be hosted on GitHub for onboarding
- /input/ directory is read-only for datasets
- Data written outside /output/ lost after job completes
- Training job must complete before fetching /output/ for large checkpoints
- Common compute pool (dedicated environment requires request)
- Additional steps needed for non-graphical environments like GitHub Codespaces
Frequently asked questions about FlexAI
What is the supported technology stack on FlexAI?
FlexAI supports Python 3.11, PyTorch 2.4, and CUDA 12.4. The available hardware is Nvidia GPUs. You should not include torch, torchvision, or torchaudio in your requirements.txt file as FlexAI will install the correct versions automatically.
What are the requirements to get onboarded to FlexAI Cloud Services?
To get onboarded, your model must be hosted on GitHub, either privately or publicly. New accounts also come with $5 to get started, and you need to register for an account to get an API key from the settings page.
Does FlexAI offer managed checkpoints and how do they work?
Yes, FlexAI offers managed checkpoints out of the box without any setup required. Every time your code calls torch.save, FlexAI stores the output so you can resume your Training Job from a previous point in case of failure. You can download checkpoints using the flexai checkpoint fetch command, even while the job is still running.
What is the maximum size for managed checkpoints?
Managed checkpoints support up to 5GB in size. If your checkpoints are larger than 5GB, the managed checkpoints feature will not be enabled, but you can continue writing to the /output/ directory and wait until the Training Job completes before fetching the contents.
How do I resume a Training Job from a checkpoint?
You can resume a Training Job by creating a new Training Job using flexai training run with the --checkpoint flag and passing the name or ID of the checkpoint you want to resume from. This works even for multi-node training jobs.
Is there a limit on dataset size for FlexAI?
FlexAI does not set a limit on the size of datasets you can use. There are also no restrictions on file types - you can upload images, text, audio, video files, tokenized data, or any other file type required for your training job.
How are dataset files and output artifacts managed?
Dataset files are made available in the /input/ directory which is read-only. Your training scripts should write output artifacts or data processing results to the /output/ directory. Data written to any other directory is lost after the Training Job completes. Once completed, /output/ contents are available for download using flexai training fetch.
How can I monitor infrastructure metrics for my training jobs?
You can use the FlexAI Infrastructure Monitor to monitor training job performance. It provides real-time system and GPU metrics to help optimize your training scripts. The data retention period for infrastructure metrics is 30 days. TensorBoard is also available for monitoring training progress.
Is the FlexAI interface CLI or GUI-based?
Today, FlexAI's interface is CLI-based, which provides developers faster system interaction, granular control, and scripting for repetitive tasks with easier integration into automated workflows. A web-based interface (Web Console) is available for managing workloads, and more will be offered soon.
Can I use FlexAI on my current infrastructure or offer it to my users?
Yes, FlexAI can support using their platform on your current infrastructure through Compute Provider partnerships. As a Compute Provider, you can also offer FlexAI as a solution to your users by leveraging your compute power. You need to reach out to FlexAI for more details on either option.