The AI control plane that runs on your infrastructure

Run everything from GPU training to agent sessions on your own clusters — scheduled, tracked, and governed in one place. Self-hosted deployments keep your data inside the infrastructure you control.

Start free

Explore the platform

$

>queued on gpu-a100 · 4 GPUs

>metrics and checkpoints streaming

Inputs

Frameworks, specs, clients, and APIs feed the platform.

ML/AIAgentsAPI / CLI+1 more

Governance

Organization, access, and policy controls admit every run.

Orgs + teamsAuthn / AuthzAudit logs+2 more

Scheduling & Orchestration

Orchestrate runs, observe systems, compare results, and register assets.

SchedulingExperimentsServing+4 more

Execution controls

Agents and queues translate product intent into cluster work.

AgentsQueuesConnections+1 more

Infra

Polyaxon runs across your clouds, clusters, accelerators, and storage.

Cloud clustersNeoCloudsOn-prem Kubernetes+3 more

InputsFrameworks, specs, clients, and APIs feed the platform.

ML/AIRuns + tracking

AgentsSchedule + trace

API / CLIClients, SDKs

PolyaxonfileYAML · JSON · Python · Go · TS

GovernanceOrgs + teamsAuthn / AuthzAudit logsPoliciesAutomation

Scheduling & OrchestrationSchedulingExperimentsServingDashboardsEvalsRegistryTemplates

Execution controlsAgentsQueuesConnectionsPresets

InfraCloud clustersNeoCloudsOn-prem KubernetesAcceleratorsStorageRegistries

3.8KGitHub stars

19.2MDocker pulls

3.7MRuns per month

1.2KCommunity members

AI-first infrastructure.

Training, services, and sandboxes share the same specs, queues, governance, and data connections — on your own Kubernetes and storage backends.

Training & Jobs

The full training loop.

Train new models, post-train open-source ones, and fine-tune with LoRA — from a single GPU to multi-node runs. Sweeps, DAGs, batch jobs, and agent executions ride the same queues and quotas.

Explore docs

Training, fine-tuning & post-training

Train models from scratch, or post-train open-source ones with LoRA and SFT — single GPU or multi-node.

Distributed & multi-node training

PyTorch DDP, MPI, and Horovod gang-scheduled across your GPU nodes from a single spec.

Parallel computing & data processing

Scale Ray, Dask, and Spark workloads for preprocessing, feature pipelines, and batch compute.

Reinforcement learning

Stateful rollouts, reward loops, and environments for long-horizon training.

Hyperparameter sweeps

Grid, random, Bayesian, and Hyperband strategies with early stopping.

DAGs & batch jobs

Multi-step pipelines, scheduled batch work, retries, and dependencies.

Agent executions

Schedule agent workloads like any run — queued, traced, in parallel, and governed.

Web scraping & media processing

Run Chromium instances — headed or headless — and process images or audio at scale.

One control plane, every feature included.

Run AI workloads across your infrastructure with tracking, observability, dashboards, logs, lineage, evaluation, automation, and governance built in.

Control Plane

Polyaxon sits between your frameworks, models, data, clusters, and hardware so engineers and agents can run training, inference, evals, notebooks, AI gateway traffic, and services from one place.Capture every detail of your experiments — hyperparameters, metrics, code versions, data lineage, and output artifacts. Compare runs side-by-side and visualize training curves in real time.Compare dense run views with charts, graphs, synced hover states, and tooltip diffs. Build dashboards, rank runs in tables, and visualize training in real time.Register artifacts, datasets, models, and prompts with shared versioning. Track lineage from data to deployed model, compare versions, and promote approved assets across teams.Observe production systems with logs, metrics, traces, health checks, and alerts. Inspect live workloads, follow lineage across runs, and debug with runtime state at hand.Run eval suites across parallel environments with reproducible snapshots, traces, metrics, and artifacts. Compare outputs in dense tables, rank results, and promote approved models.Route model calls across providers, manage credentials, enforce policies, cache responses, and trace every request — one gateway for the models your teams and agents call.Wire automation into platform state: launch evaluations when runs finish, send notifications and alerts, trigger downstream jobs with hooks and events, and run work on schedules.Control organizations, teams, roles, and access. Track activity, require approvals, and keep audit trails across every project, run, and agent session.Schedule, query, debug, and ship from the terminal. The same commands work for engineers, GitHub Actions, and agent skills driving automation.Drive the platform from the REST API or the Python and TypeScript clients. Search and filter every run, metric, and artifact with the query language, and wire results into CI/CD.

Polyaxon sits between your frameworks, models, data, clusters, and hardware so engineers and agents can run training, inference, evals, notebooks, AI gateway traffic, and services from one place. Learn more →

Any framework

PyTorch

TensorFlow

LangChain

Ray

Dask

Your data

S3

GCS

Azure Blob

NFS

Volumes

Any models

OpenAI

Anthropic

Hugging Face

Mistral

vLLM

Ollama

Polyaxon control plane for AI infrastructure

JobsServicesDAGsMatrixAgentsQueuesNamespacesPresetsSessionsEvalsGatewaysArtifacts

Any cluster

AWS

GCP

Azure

Nebius

CoreWeaveKubernetesOn-prem

Any hardware

NVIDIA GPU

AMD GPUCPU

TPU

Connections

GitHub

GitLab

Datadog

Sentry

New Relic

Slack

Docker

Control Plane

Polyaxon sits between your frameworks, models, data, clusters, and hardware so engineers and agents can run training, inference, evals, notebooks, AI gateway traffic, and services from one place. Learn more →

Any framework

PyTorch

TensorFlow

LangChain

Ray

Dask

Your data

S3

GCS

Azure Blob

NFS

Volumes

Any models

OpenAI

Anthropic

Hugging Face

Mistral

vLLM

Ollama

Polyaxon control plane for AI infrastructure

JobsServicesDAGsMatrixAgentsQueuesNamespacesPresetsSessionsEvalsGatewaysArtifacts

Any cluster

AWS

GCP

Azure

Nebius

CoreWeaveKubernetesOn-prem

Any hardware

NVIDIA GPU

AMD GPUCPU

TPU

Connections

GitHub

GitLab

Datadog

Sentry

New Relic

Slack

Docker

Works with your stack.

Track experiments from the ML libraries your team already uses, run distributed jobs on Kubernetes, route AI gateway traffic, and connect artifacts, registries, and infrastructure across cloud, on-prem, and hybrid clusters.

AI/ML tracking libraries

Hugging Faceand more...

AI agents and gateway

Rayand more...

Distributed training

Ray

RAPIDSand more...

Cloud, storage, and registries

AWS

GCP

Azure

S3

GCS

Azure Blob

NFS

PVC

MinIO

ECR

ACR

GCRand more...

Automation, services, and observability

See all integrations

Need another backend? Request it ->

Why use Polyaxon?

An opinionated AI engineering platform for teams that want full control over their infrastructure, data, execution, and governance.

Polyaxon covers ML engineering, AI training, agentic development, production monitoring, and infrastructure management in one platform.

The full AI engineering cycle
Polyaxon covers ML engineering, AI training, agentic development, production monitoring, and infrastructure management in one platform.
Built on an open-source core
Polyaxon is built on an Apache 2.0 core you can inspect, self-host, and extend through the free Community Edition.
Framework and runtime agnostic
Use PyTorch, TensorFlow, JAX, XGBoost, Scikit-learn, LLM frameworks, shell commands, Python, TypeScript, SSH, and tmux from the same control plane.
Reproducible by default
Every run can capture code, hyperparameters, dependencies, artifacts, prompts, traces, and lineage. Re-run or inspect work months later.
Scales without rewrites
Go from a single project to multi-cluster GPU fleets, agent pools, namespaces, and backend storage without changing the workflow model.
First-class CLI, SDKs, and APIs
Use CLI, REST, gRPC, Python, and TypeScript APIs to start sessions, inject code, run commands, manipulate files, use git, and call HTTP services.

Enterprise Control and Governance.

Polyaxon runs on your infrastructure - on-prem, in your VPC, or on any managed Kubernetes. Everything is designed around isolation, auditability, activity tracking, and team-level controls.

Architecture

Governance

Operations

Logs, metrics, traces, and health checks
Approvals and activity streams
Queues, agents, connections, and namespaces
Git, HTTP, filesystem, and artifact workflows

Offering

Polyaxon scales with your operations.

Polyaxon(Cloud)
Fully Managed Control Plane with 1-click deployment, orchestration, logging, backups, upgrades and security. All workload, code, data, artifacts, and models stay on your clusters.
Start free
Polyaxon(EE)
On premise Enterprise Edition Control Plane with all features, upgrades, security, monitoring, compliance, scale, and premium support.
Explore Enterprise
Open source
Running on top of Polyaxon's open-source tools, Polyaxon Community Edition is our free version with all core features to get you started.
Deploy Community

Questions & Answers

Polyaxon is an open-source AI engineering control plane for ML workloads, AI training, agent sandboxes, coding sessions, observability, and model management on Kubernetes. Try it with the community edition or Polyaxon Cloud.

Ready to give it a try?

Start with Polyaxon Cloud or deploy the Community Edition for free.

Get started

Learn more!

Get in touch to learn more about our Enterprise offering.

Request a demo

The AI control plane that runs on your infrastructure

AI-first infrastructure.

The full training loop.

One control plane, every feature included.

Control Plane

Control Plane

Works with your stack.

Why use Polyaxon?

The full AI engineering cycle

Built on an open-source core

Framework and runtime agnostic

Reproducible by default

Scales without rewrites

First-class CLI, SDKs, and APIs

Enterprise Control and Governance.

Offering

Polyaxon(Cloud)

Polyaxon(EE)

Open source

Questions & Answers

?What is Polyaxon?

?What does Polyaxon help me with?

?What frameworks are supported?

?What deployment options are available?

?Is Polyaxon actually open source?

?How does it scale?

?Is Polyaxon secure and compliant?

?How do I get started?

?How does pricing work?

Get started with Polyaxon

Ready to give it a try?

Learn more!

Control Plane

Control Plane