Is there a comprehensive list of Outerbounds' features anywhere? • integrations • bakery • workstations • considerations for alerting • consideratiosn for branching • etc. We'll do an evaluation but I'd like to have some initial arguments of why to use it over SageMaker, Airflow, and ZenML. We're about to start a round of evaluations.

not a comprehensive list, but here are some highlights 🙂 ✨ Data • Snowflake ◦

@snowflake

/

@snowpark

• Fast Data Compute • Bring your own cloud • Multi-cloud ◦ GCP / Azure / AWS / CoreWeave etc • Cost reporting / optimization • Cloud GPUs ◦

@nvidia

• Distributed training / inference ◦

@checkpoint

/

@model

• Automatic containerization with Fast Bakery •

@slurm

LLM / GenAI • Finetune with

torchtune

/

llamafactory

/ HF • NVIDIA

@nim

◦ private LLM endpoints ◦ NVIDIA Enterprise AI license included Orchestration • Event bus / triggering • Highly available production scheduling • Large-scale scheduling ◦ Up to tens of thousands of parallel tasks • Notify on Pagerduty / Slack • Visualize interconnected flows (coming soon!) Development • Workstations ◦ Web IDE ◦ VSCode extension • Visualizations / observability ◦ Real-time cards ◦

@datadog

/

metaflow.measure

• Authenticated apps ◦ Streamlit etc • CI/CD integration • Experiment tracking ◦ Metaflow artifacts ◦

@tensorboard

Compliance / Security • All user auth via SSO • Machine tokens for server-to-server auth • Secrets manager ◦ Built-in / bring-your-own • Secure perimeters • Bring Your Own Policy • SOC2 • HIPAA • Global / regional deployments

Thank you thank you thank you @victorious-lawyer-58417 ! And, I remember that the step functions deployment type doesn’t have as much support now as Argo workflows. If we want to be outerbounds customers, but our group is rigid on deploying to Airflow, is there anything we lose? (Anything loss in capabilities, support, DexEx, etc.)

you can definitely do both. You wouldn't be the first shop where DEs want to run legacy pipelines on Airflow. Eventing works well both ways

I wouldn’t even say they’re legacy pipelines. I think in his mind this is the future: all teams on one tool.

that’s not a bad vision but realistically open-source Airfow is not the tool for the job. Many companies have tried and ended up writing their own solution because of widely documented issues with Airflow

I am so using this quote 😆