Union.ai
Union.ai is a Seattle startup whose managed Flyte runs a product’s AI training, data and agent pipelines in its own cloud, with startup discounts.
Founder Stack
Listed in AI
- Contact
Is this your organization? Claim it
Facts
- HQ
- Seattle, WA, US
- Sectors
- Software Development
- Founded
- 2021
- Employees
- 69
- Stage
- Series A
Sage Elliott: Flyte Lets You Deploy Streamlit, Gradio, and FastAPI Apps Next to Pipelines Using AppEnvironment
Sage Elliott explains how Flyte lets you deploy Streamlit, Gradio, and FastAPI apps alongside pipelines using AppEnvironment so dashboards, demos, and APIs share images, resources, auth, and can scale to zero.
Sage Elliott announces Flyte MCP server and 23 agent skills to let Claude, Codex run and auto-fix Flyte 2 pipelines
Sage Elliott highlights Flyte's new MCP server and 23 agent 'skills' that let coding agents (Claude, Codex, etc.) run, debug, and fix Flyte 2 pipelines automatically, removing the manual copy‑paste loop when tasks fail.
Kyle Hazen shows an agent on Flyte 2 complete a broken RPA price-update workflow after supplier portal redesign, proving agents endure UI changes where RPA fails
Kyle Hazen demonstrates replacing brittle RPA bots with goal-directed agents by rebuilding a real distributor price-update workflow on Flyte 2, intentionally breaking the supplier portal redesign, and logging every decision the agent made as it completed the task despite UI changes.
Pandera 0.36 introduces ParsedColumn/ParsedField and pandera.decisions to declare and populate derived columns
Pandera 0.36 adds ParsedColumn and ParsedField to declare how a column is derived and a new pandera.decisions module to populate typed columns by querying a decision model, letting schemas track provenance, costs, and validation of derived data.
Sage Elliott announces Flyte new_condition API lets tasks pause for human approvals to gate deploys and actions
Sage Elliott explains how Flyte's new_condition API lets tasks pause and present human approval prompts (rendered in markdown) that survive restarts and wait without polling, enabling safe gates for actions like production deploys or refunds.
Union.ai’s Sergey Vilgelm announces resource-aware scheduling that pre-checks cluster capacity and usage to avoid stuck GPU jobs
Union’s scheduler gained resource-aware scheduling so it checks cluster capacity and usage before dispatching jobs, avoiding indefinite Pending pods by failing impossible requests and routing tasks to the least-loaded suitable cluster.
Niels Bantilan explains how Union Artifacts add named, versioned outputs, partitions, and triggers to build durable, event-driven AI pipelines
The article explains how Union Artifacts provide durable, versioned names for outputs so consumers can reliably find data, and extends that model with partitions for sliced data and triggers to start downstream work as soon as new versions arrive.
Kyle Hazen announces nf-flyte plugin lets nf-core Nextflow pipelines run unchanged on Union as Flyte actions
nf-flyte is a Nextflow executor plugin that lets you run existing nf-core pipelines on Union without changing them, turning each Nextflow process into a Flyte action so outputs flow directly into Python and ML tasks.
Niels Bantilan announces Union self-serve supports Google Cloud: connect a GKE cluster to run training, pipelines, agents, and inference inside your project
Niels Bantilan announces that Union self-serve now supports Google Cloud, letting teams connect a GKE cluster in their own project to run training, pipelines, agents, and inference while keeping data, traces, and compute inside their perimeter. This enables organizations already on GCP to use Union without migrating infrastructure, addressing production-line orchestration and failure modes that infrastructure alone doesn't solve.
Niels Bantilan: Union Artifacts defined as named, versioned references with provenance to avoid duplicate storage and registries
Niels Bantilan explains that artifacts are named, versioned references to objects in object storage that include provenance and dependency records, letting teams manage datasets and models without duplicating data or creating separate, inconsistent registries.
Sage Elliott: Flyte 2.12 Enables Native Apple M‑series GPU Tasks via gpu="AppleSilicon" in Devbox
Sage Elliott explains that Flyte 2.12 lets developers run GPU-accelerated Flyte devbox tasks on Apple M-series Macs by requesting gpu="AppleSilicon", which runs task code natively on the host using PyTorch's mps and MLX while the devbox still manages inputs, outputs, and logs.
Niels Bantilan announces Flyte Devbox adds Apple Silicon GPU support so gpu="AppleSilicon" tasks run natively on Mac
Niels Bantilan explains that Flyte 2's devbox now supports an Apple Silicon accelerator so tasks requesting gpu="AppleSilicon" run natively on Macs, enabling torch MPS and MLX to access the Mac GPU for training and serving. The implementation runs workload code on the host macOS while the container acts as a proxy, avoiding Metal-in-Linux limitations and preserving the familiar Flyte devbox API.
People
-
Ketan Umare Co-founder and CEO