paguasmar
VeloxML aims to make production LLM infrastructure simple to deploy while keeping data, prompts, and model weights inside a customer's own AWS or GCP account. Its open-source CLI provisions and scales GPU-backed model endpoints with one command, without Dockerfiles, Kubernetes YAML, or proprietary framework decorators.
At a Glance
- AI and ML engineering teams
- Organizations serving open-source or fine-tuned LLMs
- Regulated or privacy-sensitive companies needing data sovereignty
- Teams with AWS or GCP accounts and private VPC requirements
- +1 more
AI Tools by paguasmar
(1)
VeloxML
Open Source LLM Deployment CLI
Discussions
No discussions yet
Be the first to start a discussion about paguasmar
Latest News
README documentation cleaned up to remove stray benchmark notes
VeloxML v0.1.1 released with rolling updates, autoscaler fixes, and CLI DX improvements
PyPI publish workflow added
Initial v0.1.0 release introduced the VeloxML CLI and deployment engine
Products & Services
An Apache-2.0 open-source Python CLI and deployment engine for deploying open-source LLMs and other AI services directly into a customer's AWS or GCP account. It reads model code and a veloxml.yaml hardware specification, provisions Spot or On-Demand compute through SkyPilot, installs dependencies, runs health checks, and returns a live inference endpoint.
Market Position
VeloxML positions itself against Modal, RunPod, and Baseten as a self-hosted, cloud-account-native alternative. Its stated differentiators are keeping prompts, data, and model weights in the customer's AWS/GCP account, avoiding enterprise private-VPC paywalls and third-party compliance reviews, using Spot instances for cost efficiency, and avoiding proprietary decorators or framework lock-in.
Leadership
Founders
Pedro Águas Marques
Senior AI and MLOps engineer; LinkedIn search results describe him as the sole operational owner of four ML services in the critical path of a $300M+ ARR product and as ex-Tripadvisor. He is the GitHub owner and sole contributor shown for veloxml-deploy, and the package metadata names him as author.
Executive Team
Pedro Águas Marques
Founder and maintainer
Senior AI and MLOps engineer and ex-Tripadvisor; his public profile describes operational ownership of four ML services in a critical-path $300M+ ARR product. He is the GitHub owner, package author, and sole listed contributor for VeloxML.
Founding Story
Pedro Águas Marques said he originally wrote VeloxML to replace Modal and Baseten because he wanted to deploy open-source models inside his own VPC without vendor lock-in or proprietary Python decorators. The initial vision was one-command deployment and scale-to-zero on customer-owned infrastructure, consolidating LLM serving while reducing cost and keeping data within the customer's security perimeter.
Business Model
Revenue Model
The publicly documented project is open-source under Apache 2.0 and does not publish a paid software subscription or license plan. Users bring their own AWS/GCP accounts and pay the underlying cloud-compute costs; VeloxML's positioning is to avoid third-party cloud markup.
Pricing Tiers
The CLI is Apache-2.0 licensed. The README gives an example of deploying Qwen2.5-0.5B on an AWS Spot instance at approximately $0.07/hour; a Hacker News reply says the controller can cost up to $0.08/hour and larger model costs depend on the selected instance.
Target Markets
- AI and ML engineering teams
- Organizations serving open-source or fine-tuned LLMs
- Regulated or privacy-sensitive companies needing data sovereignty
- Teams with AWS or GCP accounts and private VPC requirements
- Developers seeking an alternative to managed inference platforms and vendor lock-in
- Private LLM inference APIs
- Custom embedding and reranking microservices
- Real-time reasoning and agent tool-calling backends
- Secure hosting of domain-specific fine-tuned models inside a private VPC
- High-throughput batch inference endpoints
- Production model serving where data sovereignty, compliance, or cloud-cost control is important