Nodus Compute
Nodus is an AI compute platform that matches containerized workloads—such as training, fine-tuning, and agent inference—with compatible GPU capacity and manages the entire run lifecycle. It provides managed runtimes for post-training methods like SFT, DPO, and GRPO, and integrates with coding agents like Claude Code, Codex, and Cursor via an MCP server. The platform enforces granular spending caps and blocking budgets to prevent runaway agent costs.
- Artificial Intelligence
- Developer Tools
- Software Only
Funding
Raised to date
$500KRaised to date based on public sources. This may differ from the amount the company actually raised and is based only on what is publicly available on the internet.
Founders
Founder details are not available yet.
Product
Problem
Developers and AI engineers running training, fine-tuning, and inference workloads struggle to find compatible GPU capacity, manage infrastructure around their jobs, and control costs when using coding agents that can launch compute automatically. Existing cloud platforms lack granular, enforceable budget controls that stop runaway agent spend in real time, and recovering from spot-preemption or interrupted runs requires complex manual orchestration.
Solution
Nodus is an AI compute platform that matches containerized workloads—such as training, fine-tuning, and agent inference—with compatible GPU capacity and manages the entire run lifecycle. The platform handles infrastructure automatically, from matching workloads to compatible capacity, estimating costs before runs begin, executing jobs, and recovering from interruptions. For training specifically, Nodus provides managed runtimes for SFT, DPO, ORPO, KTO, GRPO, distillation, and evaluation, handling model caching, checkpointing, and validation. The platform also integrates with coding agents like Claude Code, Codex, and Cursor through an MCP server, enabling agents to request compute, set spending limits, follow logs, and retrieve results without leaving their coding environment.
Target Audience
Primary customers are AI engineers, ML researchers, and development teams using coding agents who need GPU capacity for training, fine-tuning, and inference workloads, as well as organizations seeking reliable, cost-controlled infrastructure for post-training workflows.
Features
- Managed training runtimes covering supervised fine-tuning (full or LoRA/QLoRA), DPO, ORPO, KTO, reward modeling, GRPO, distillation, and evaluation
- Enforceable per-run spending caps (`maxCostUSD`) and blocking budgets scoped to projects, members, or labels, with action:Block that stops work before additional costs accrue
- MCP integration for Claude Code, Codex, Cursor, and other agents, allowing them to launch jobs, estimate costs, set limits, and retrieve outputs from within the coding session
- Atomic checkpointing with state directories for recovery from spot preemption, including automatic checkpoint requests and replacement machine provisioning
- Validation-based checkpoint selection with configurable validation fractions, patience, and improvement thresholds
- Dry-run estimates for training jobs, showing cost-to-completion and resource allocation before any paid work begins