Union
Union.ai provides Flyte, an open-source workflow orchestration platform that manages the execution, scheduling, and monitoring of ML training jobs, model serving, and AI agent workloads across distributed Kubernetes clusters and multiple cloud providers. The platform includes a durable runtime that guarantees task completion even if infrastructure fails, an inference gateway for routing API requests to self-hosted or third-party models, and priority-based scheduling with quotas to allocate compute resources fairly. Union integrates with PyTorch, Hugging Face, Kubeflow, and MLX, allowing teams to orchestrate existing ML pipelines without rewriting code. Pricing is usage-based (per action, per GPU-hour, per vCPU-hour) or fixed-plan (Team plan at $950 USD per month), with add-ons for managed clusters and guided installation.
Union is an AI infrastructure platform, priced at $950 a month on the Team plan, integrating with AWS Marketplace, Hugging Face, PyTorch and Kubeflow. InnovaAI rates it 2.6 of 10 for agency adoption, best for ML Engineer, Infrastructure Engineer and DevOps Lead roles.
Agency Audit
Union.ai provides Flyte, an open-source workflow orchestration platform that manages ML training, model serving, and AI agent deployment across distributed clusters and clouds. For digital agencies building or deploying AI-powered client solutions, Union is relevant only if your team operates an in-house ML infrastructure practice, trains custom models, or manages inference pipelines at scale. Most traditional creative and strategy agencies will not benefit from adoption; this tool is built for ML engineering and AI infrastructure teams, not general agency operations.
3recommended
36/mo
$1,750/mo
High
Illustrative scenario. Not a guarantee. Net capacity is the value of reclaimed time at $75/hr, less the lowest verified paid base plan (flat plan cost is shared). Hours saved come from the service estimate; implementation, taxes, and unprovided usage charges are excluded.
- ML Engineer handling model training and fine-tuning orchestration
- Infrastructure Engineer handling multi-cloud inference endpoint management
- DevOps Lead handling distributed workload scheduling and failover
- Your agency does not operate in-house ML infrastructure or does not train custom models. Union is an infrastructure tool, not a client-facing or general productivity platform.
- Your team uses managed ML services like SageMaker, Vertex AI, or Hugging Face Inference API and does not need multi-cloud orchestration. Union adds complexity without offsetting benefit.
- You have fewer than 2 full-time ML engineers or infrastructure staff. Union's operational overhead requires dedicated DevOps expertise; smaller teams will spend more time troubleshooting than saving.
Internal Adoption Path
$950/mo
$950/mo flat plan
36 hr/mo
3 seats × 12 hr each
$2,700/mo
modeled at $75/hr labor rate
$1,750/mo
value − subscription cost
In this model, 3 seats reclaim 36 hours of team time each month. Valued at $75/hr that is $2,700/mo, and after the $950/mo subscription it leaves $1,750/mo of capacity for billable client work.
Illustrative scenario. Not a guarantee. Uses the lowest verified paid base plan. Implementation, taxes, and unprovided usage charges are excluded.
Platform Features
Core capabilities of Union
Distributed ML workflow orchestration
Automates scheduling and execution of training jobs, data pipelines, and inference tasks across multiple clusters and cloud providers. ML engineers save 3-5 hours per week on manual job submission, monitoring, and cross-cluster failover.
Durable runtime with fault tolerance
Guarantees task completion even if nodes fail or infrastructure restarts. Eliminates manual job resubmission and reduces infrastructure team's incident response time by consolidating retry logic into the platform.
Inference gateway and model serving
Centralizes API routing, load balancing, and model versioning across self-hosted and third-party providers. Reduces infrastructure team's time spent on manual endpoint management and provider failover.
Priority-based scheduling and quotas
Allocates compute resources by workload priority and enforces per-team or per-project limits. Prevents resource contention and gives infrastructure teams visibility into cost allocation per client or project.
Multi-cloud and multi-region deployment
Routes workloads to AWS, GCP, Azure, or on-premise clusters based on cost, latency, or availability. Eliminates vendor lock-in and lets infrastructure teams optimize spend across cloud providers.
Native Apple Silicon GPU support via Flyte devbox
Enables ML engineers to develop and test workflows locally on M-series Macs before deploying to production clusters. Reduces iteration time for model development and testing.
What Makes Union Different
Unique advantages vs similar tools in this niche
Durable runtime recovers runs from infrastructure failure instead of restarting
vs Standard orchestration tools that restart failed pipelinesRuns that recover from infrastructure failure instead of restarting.
Native Apple Silicon GPU support in the devbox
vs Linux-only container devboxes without Metal supportA devbox task or app that asks for gpu="AppleSilicon" now runs natively on your Mac.
Value Equation
Outcome-likelihood-time-effort assessment for Union
Limited agency channel
Union scored below the agency-resellability threshold (agency_fit_score < 50). The Value Equation projects agency-side outcomes, which don't apply to tools without a clear resell pathway.
Contact UnionPricing
Union platform cost to your agency
Team: $950/mo
Team
- 2 active clusters
- 10,000 concurrent actions
- 200,000 actions per run
- 30-day data retention
Enterprise
- Significant volume discounts on every meter
- Unlimited clusters, multi-region and multi-cloud
- Custom concurrency, fanout and retention
- Zero Trust security and observability
How usage-based pricing works
Union charges per consumption unit (per action (10m +)). Below are the component rates the vendor publishes. Each row is a separate charge: your total cost combines them based on your configuration and volume. Component rates range from $0.0003 per action (10m +).
Final agency cost = (sum of selected component rates) × client usage volume. Confirm a usage estimate with each client before quoting.
Component Rates
Cost per unit: total depends on your configuration and volume
Add-ons
Optional extras priced on top of any main plan
No verified white-label program for Union: client-facing delivery runs under the platform's native branding.
Market Intelligence
Offer + scale economics for Union
Limited agency channel
Union scored below the agency-resellability threshold (agency_fit_score < 50). It's a useful tool but not designed for white-labeled or retainer-based reselling, so we don't publish productized offer economics for it.
Contact UnionInvestment Decision Framework
Strategic vetting analysis for Union
Skip
Weak agency-resell fit
Buy If
4Your ML engineering team trains and fine-tunes custom models for client projects 10+ hours per week and currently manages orchestration manually or via ad-hoc scripts. Union's distributed training and scheduling would consolidate that workflow and reduce failed job recovery time.
Your infrastructure team manages inference endpoints across multiple cloud providers and needs centralized governance of API routing and model serving. Union's inference gateway and multi-cloud scheduling eliminate manual failover and quota management.
Your team deploys AI agents that require durable runtime and fault tolerance across clusters. Union's runtime guarantees and priority-based scheduling prevent dropped tasks during infrastructure churn.
You work with biotech, robotics, or life sciences clients and operate internal ML pipelines for model validation or data processing. Union's native support for distributed workloads and GPU scheduling reduces infrastructure provisioning friction.
Skip If
5Your team uses managed ML services like SageMaker, Vertex AI, or Hugging Face Inference API and does not need multi-cloud orchestration. Union adds complexity without offsetting benefit.
Your agency does not operate in-house ML infrastructure or does not train custom models. Union is an infrastructure tool, not a client-facing or general productivity platform.
You have fewer than 2 full-time ML engineers or infrastructure staff. Union's operational overhead requires dedicated DevOps expertise; smaller teams will spend more time troubleshooting than saving.
Your ML workloads are small, infrequent, or one-off. Union's pricing model (per action, per GPU-hour, per vCPU-hour) and setup cost ($5,000 to $15,000 for guided install) only pay back at scale.
Your team lacks Kubernetes or cloud infrastructure experience. Union requires cluster management and networking knowledge; adoption without that expertise will stall.
Bottom Line
Union.ai provides Flyte, an open-source workflow orchestration platform that manages ML training, model serving, and AI agent deployment across distributed clusters and clouds. For digital agencies building or deploying AI-powered client solutions, Union is relevant only if your team operates an in-house ML infrastructure practice, trains custom models, or manages inference pipelines at scale. Most traditional creative and strategy agencies will not benefit from adoption; this tool is built for ML engineering and AI infrastructure teams, not general agency operations.
Reality Check
Union requires deep infrastructure expertise to deploy and operate. Setup involves cluster configuration, Kubernetes knowledge, and ongoing DevOps overhead. Payback only materializes if your agency runs 5+ concurrent ML workloads or trains models regularly; one-off AI projects do not justify the operational complexity.
High effort: requires technical configuration and team training
Academy for Union
Work through it in order: the course for this service first, then the modules behind it.
No Academy modules are published for this service yet. Browse the full Academy
Why this category matters
The commercial case before the tooling.
Core concepts
The mental model you need to price and scope the work.
- Concentration Risk LedgerConcept
Concentration Risk Ledger is a framework for tracking how much of an agency's delivery capacity depends on any single model provider, region, or price tier. The unit of analysis is not the vendor relationship but the retainer: for each client engagement, list which workflows break if one provider raises prices, degrades quality, or restricts access. Forrester warned in October 2026 that AI supply chains hide single points of failure in plain sight, and the same week Anthropic cut Claude Haiku 5.5 to $0.10 per million input tokens while OpenAI shipped GPT-6 to 1.2 billion weekly users, both reminders that pricing and capability floors move fast. An agency running every client summarization job through one API has an unpriced liability. The ledger converts that into a number: percentage of monthly delivery hours exposed, and the cost of a routing layer that reduces it.
- Provider Concentration RatioConcept
Provider Concentration Ratio is the share of an agency's live client delivery that depends on a single model vendor, measured in billable projects rather than API keys. A retainer running three client builds on one provider's API looks diversified on an invoice and is not: a pricing change, a deprecation, or a guardrail shift hits all three at once. The ratio matters because infrastructure cost is a pass-through that agencies rarely reprice mid-contract, so a vendor move lands directly on margin. The practical test is to count distinct providers behind each deliverable and flag any client where one vendor carries more than half the inference load. Routing layers such as OpenRouter, Portkey, and Helicone exist to make that count adjustable without rewriting application code. The same discipline applies to compute: industry executives expect only 7 to 10 of hundreds of AI cloud providers to survive consolidation, so GPU capacity booked with a single neocloud carries the same concentration risk as a single-model stack.
- Inference Cost FloorConcept
Inference Cost Floor is the practice of tracking the lowest available price per million tokens for a capability tier, then treating every drop as a trigger to re-price client retainers rather than a windfall to bank. Agencies that price AI work on today's model economics get undercut the moment a cheaper tier ships, because the client's procurement team reads the same launch posts. Anthropic's Claude Haiku 5.5 arrived at $0.10 per million input tokens with a 1 million token context window, which resets what high-volume document summarization and campaign analysis should cost a client. The framework has three moves: benchmark your current blended cost per deliverable, set a review cadence tied to model releases, and pre-agree with clients that savings split rather than vanish. Agencies running fixed-fee AI retainers without a floor review are quietly donating margin every quarter.
Decision and risk
How to judge the fit, and the ways it goes wrong.
- AI Infrastructure Rule: Price the Exit Before You Price the InferenceEvaluation Rule
Treat provider portability as a delivery requirement: put a gateway or routing layer in front of every model call, and price the migration path into the retainer before the first token is billed.
- AI Infrastructure Rule: Route Through a Gateway Before You Sign a Second Model ContractEvaluation Rule
Stand up a gateway layer that logs, caches, and routes requests across at least two providers before any single-model contract becomes load-bearing for a client retainer.
- The Single-Provider Trap: Why AI Infrastructure Stalls When One Model Vendor Owns the StackFailure Pattern
- The Token Bill Trap: Why AI Infrastructure Costs Outrun Agency RetainersFailure Pattern
8 modules selected for Union
Frequently Asked Questions
Answers about pricing, setup, implementation
Union.ai provides Flyte, a workflow orchestration platform that schedules and executes ML training, model serving, and AI agent workloads across distributed clusters and multiple cloud providers. It includes an inference gateway for routing API requests to self-hosted or third-party models, durable runtime guarantees to prevent task loss, and priority-based scheduling to allocate compute resources fairly. Integrations with PyTorch, Hugging Face, and Kubeflow allow teams to orchestrate existing ML pipelines without rewriting code.
Union offers usage-based and fixed-plan pricing. The Team plan costs $950 USD per month and includes 2 active clusters, 10,000 concurrent actions, and 200,000 actions per run. Per-action pricing ranges from $0.003 USD (0–100k actions per month) to $0.0003 USD (10M+ actions per month). GPU-hour pricing ranges from $0.068 USD (Class 1, 0–1k hours per month) to $0.59 USD (Class 3, 10k+ hours per month). Add-ons include Union-managed cluster ($500 USD per month), guided install ($5,000 to $15,000 USD one-time), and single-tenant control plane ($50,000 USD per year for Enterprise). Enterprise plans include unlimited clusters and custom pricing; contact sales for a quote.
ML engineers and infrastructure teams benefit most. ML engineers compress model training and fine-tuning workflows by delegating orchestration and scheduling to Union instead of managing jobs manually. Infrastructure teams reduce operational overhead by consolidating multi-cloud resource allocation, failover, and quota enforcement into a single platform. DevOps staff spend less time on incident response when Union's durable runtime prevents task loss. Founders and operations leaders gain cost visibility through per-project quota tracking and multi-cloud spend optimization.
For ML engineering teams running 5+ concurrent training or inference workloads, Union saves 3-5 hours per week by automating job orchestration, retry logic, and cross-cluster failover. Infrastructure teams managing multi-cloud deployments save 2-4 hours per week on manual endpoint management and provider failover. Savings scale with workload volume; teams running fewer than 2 concurrent jobs see minimal time recovery and should not adopt.
Yes. Union is built on Kubernetes and requires cluster management, networking, and cloud infrastructure knowledge to deploy and operate. Teams without in-house DevOps or infrastructure staff will struggle with setup and troubleshooting. Union offers guided install services ($5,000 to $15,000 USD one-time) to accelerate onboarding, but ongoing cluster management remains the team's responsibility.
Initial cluster setup and configuration typically takes 2-4 weeks for teams with Kubernetes experience. Integrating existing ML pipelines into Union workflows adds 1-2 weeks per pipeline. Teams without infrastructure expertise should budget 6-8 weeks and consider the guided install service to reduce friction. Rollout time scales with team size and pipeline complexity.
Flyte is open-source, so workflows and job definitions remain portable. However, any data stored in Union's managed infrastructure or control plane will be inaccessible after cancellation unless you export it beforehand. Teams should plan data migration and backup strategies before adopting, especially if using Union-managed clusters or single-tenant control plane add-ons.
Union integrates with PyTorch, Hugging Face, Kubeflow, and MLX, allowing teams to orchestrate existing model training and inference code without rewriting pipelines. It also connects to AWS Marketplace for cloud resource provisioning. If your team uses other ML frameworks or custom infrastructure, you may need to build custom adapters or use Union's API to integrate them.